universidade federal fluminense programa de pÓs...
TRANSCRIPT
UNIVERSIDADE FEDERAL FLUMINENSE
PROGRAMA DE PÓS GRADUAÇÃO EM MEDICINA VETERINÁRIA
ÁREA DE CONCENTRAÇÃO: HIGIENE VETERINÁRIA E
PROCESSAMENTO TECNOLÓGICO DE PRODUTOS DE
ORIGEM ANIMAL
LEONARDO VARON GAZE
ESTUDO SENSORIAL E FÍSICO-QUÍMICO EM DOCE DE
LEITE PASTOSO
NITERÓI
2014
LEONARDO VARON GAZE
ESTUDO SENSORIAL E FÍSICO-QUÍMICO EM DOCE DE LEITE PASTOSO
Dissertação apresentada ao Programa de Pós Graduação em Medicina Veterinária da Universidade Federal Fluminense como requisito parcial para obtenção do Grau de Mestre. Área de concentração: Higiene Veterinária e Processamento Tecnológico de Produtos de Origem Animal.
ORIENTADORA: Profa. Dra. MONICA QUEIROZ DE FREITAS
CO-ORIENTADOR: Prof. Dr. ADRIANO GOMES CRUZ
Niterói
2014
LEONARDO VARON GAZE
ESTUDO SENSORIAL E FÍSICO-QUÍMICO EM DOCE DE LEITE PASTOSO
Dissertação apresentada ao Programa de Pós Graduação em Medicina Veterinária da Universidade Federal Fluminense como requisito parcial para obtenção do Grau de Mestre. Área de concentração: Higiene Veterinária e Processamento Tecnológico de Produtos de Origem Animal.
BANCA EXAMINADORA
Profa. Dra. Monica Queiroz de Freitas - Orientadora Universidade Federal Fluminense
Prof. Dr. Adriano Gomes Cruz Instituto Federal do Rio de Janeiro
Prof. Dr. Carlos Adam Conte Júnior Universidade Federal Fluminense
Niterói 2014
Dedico este trabalho à minha família e amigos, em especial aos meus pais Miriam S.V. Gaze e Reinaldo V. Gaze pelo apoio que nunca faltou.
AGRADECIMENTOS
Aos meus pais, Miriam Suzi Varon Gaze e Reinaldo Gaze pelo apoio, preocupação
e atenção que sempre procuraram ter em todos os momentos vividos até hoje. Da mesma forma à minha irmã Marcelle Varon Gaze por, de formas bem particulares,
manifestar seus anseios e desejos com relação ao eu bem. À Marcela de Figueiredo Silva, por sempre estar pronta a me receber de braços
abertos e a dar conselhos totalmente aplicáveis. Também, por ter me servido de inspiração através do seu perfil profissional (ética e competência). Difícil expressar apenas com palavras sobre sua importância. À minha prima Andrea Gina Varon e à minha tia Rosangela Gaze. Médicas,
pesquisadoras, que me ofertaram momentos descontraídos, incentivo, companheirismo e que sempre se colocaram dispostas a me ajudar. Aos meus amigos de música e de vida fora dos ambientes de trabalho da UFF, Aline Moreno, Thays Hugo, Francisco Correa e Waldenir Duarte. Muito
presentes nos momentos descontraídos em que a música e assuntos extra acadêmicos deixam a alma mais leve. Às minhas “irmãs” de orientação Bruna Rosa de Oliveira e Fernanda Torres Romano por toda sua preocupação, por sempre tentarem me dar a mão quando
preciso e por me darem lições que nem sempre são boas de se receber, mas que devem ser acatadas. Às amigas Beatriz da Silva Frasão, Jesieli Braz Frozi, Marion Pereira da Costa e Bruna Leal Rodrigues, sempre juntas a mim, mesmo que à distância, muitas vezes
encarando a vida de forma parecida à como eu encaro. Presentes que ganhei da UFF. Também deixo um agradecimento especial ao amigo Eduardo Bruno Nogueira pelo apoio e boa vontade em sempre tentar ajudar ao próximo. Aos amigos Hugo Leandro Azevedo da Silva, André Muniz Afonso, Viviane da Silva Gomes, Nathalia Coutinho, Guilherme Sicca Lopes Sampaio, Karoline Ribeiro Palmeira, Carolina Cristina Colão Barcellos, Celso Fasura Balthazar, Vitor Luiz de Melo Silva, Mariana Bacellar Ribas Rodrigues, Letícia Fraga Matos Campos de Aquino e Raphael Ferreira Barros que tanto me fizeram rir e que
tornaram essa jornada tão agradável. Pena que dois anos passam voando. Agradeço aos amigos que conheci através pós-graduação em Ciência dos Alimentos da UFRJ, Ellen Cistina Quirino Lacerda, Denize Cristine Rodrigues de Oliveira, Hugo Rangel Fernandes e Eveline Kássia Braga Soares por sempre que possível se colocarem dispostos e ajudar e, mesmo com as distâncias e tarefas do cotidiano, não deixarem a distância nos afastar. Agradeço às minhas primeiras orientadoras da época da graduação na UFF, professoras Virgínia Léo Almeida e Dayse Abreu, por me introduzirem a esse
ambiente universitário “extra” salas de aula e pelos exemplos que me são até os dias de hoje. Ao meu ex-orientador, grande amigo e eterno companheiro professor Luciano Antunes Barros por sempre querer meu bem e, através de palavras precisas e nas
horas mais necessárias, saber me motivar. Também agradeço por todas as risadas que, por certas vezes, me fizeram perder a respiração. Agradeço à professora e amiga Claudia Emília Teixeira, que sempre me incentivou a fazer primordialmente aquilo que me fará feliz.
Agradeço ao professor Carlos Adam Conte Júnior e ao professor Adriano Gomes Cruz pela experiência transferida, dedicação prestada a mim, pelos puxões de orelha e palavras motivadoras, todos importantes para que o trabalho pudesse ser bem executado. Gostaria de deixar um agradecimento especial à minha orientadora, professora Mônica Queiroz de Freitas. Um grande exemplo de mãe, orientadora, amiga e de profissional. Uma pessoa que possui um seu jeito bem humorado de levar a vida e o trabalho demonstrando satisfação pelo que faz. Me trouxe paciência, serenidade e segurança durante os momentos de maior preocupação. Não tenho como me arrepender do dia em que tive a oportunidade de escolhê-la como orientadora. Agradeço à equipe da secretaria de pós-graduação, Drausio de Paiva Ferreira, Mariana F. de Oliveira e André Luis da Silva Veiga pela paciência, atenção e carinho e por muitas vezes “facilitarem as nossas vidas”, e aos coordenadores da pós-graduação, professores Sérgio Borges Mano e Eliane Teixeira Mársico, pela infraestrutura necessária ao bom funcionamento do programa. A todos os professores da pós-graduação pelo tempo e ensinamentos concedidos. Ao Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq),
pelo apoio financeiro.
“Dizem que a vida é para quem sabe viver,
mas ninguém nasce pronto. A vida é para
quem é corajoso suficiente para se arriscar e
humilde o bastante para aprender.
Clarice Lispector
SUMÁRIO
RESUMO, p. 9
ABSTRACT, p.11
1 INTRODUÇÃO, p. 13
2 FUNDAMENTAÇÃO TEÓRICA, p. 15
2.1 DOCE DE LEITE PASTOSO, p. 15 2.2 ANÁLISES FÍSICO-QUÍMICAS, p. 15 2.3 ANÁLISES SENSORIAIS, p. 18 3 DESENVOLVIMENTO, p. 19
3.1 ARTIGO 1: OPTIMIZATION OF THE DULCE DE LECHE PROCESSING: CONTRIBUTIONS OF THE SENSOMETRIC TECHNIQUES, p. 19 3.2 ARTIGO 2: DULCE DE LECHE, A TYPICAL PRODUCT OF LATIN AMERICA: CHARACTERIZATION BY PHYSICOCHEMICAL, OPTICAL AND INSTRUMENTAL METHODS, p. 59 4 CONSIDERAÇÕES FINAIS, p. 95
5 REFERÊNCIAS BIBLIOGRÁFICAS, p. 96
6 APÊNDICES, p. 106
6.1 CONFIRMAÇÃO DE SUBMISSÃO DO ARTIGO INTITULADO: OPTIMIZATION OF THE DULCE DE LECHE PROCESSING: CONTRIBUTIONS OF THE SENSOMETRIC TECHNIQUES, p 106 6.2 CONFIRMAÇÃO DE SUBMISSÃO DO ARTIGO INTITULADO: DULCE DE LECHE, A TYPICAL PRODUCT OF LATIN AMERICA: CHARACTERIZATION BY PHYSICOCHEMICAL, OPTICAL AND INSTRUMENTAL METHODS, p. 107
RESUMO
O Doce de Leite (DL) constitui-se de leite concentrado e sacarose, sendo produzido
tipicamente na América do Sul. O objetivo deste trabalho foi conhecer as
características sensoriais, físicas e químicas do Doce de Leite pastoso. A partir de
sete marcas tradicionais do mercado, obtidas em suas embalagens originais, foram
realizados testes sensoriais de aceitação empregando a escala hedônica, escala do
ideal (JAR) e de intenção de compra. As amostras também foram submetidas à
análise descritiva quantitativa (ADQ), e diferentes modelos estatísticos foram
empregados para avaliar as semelhanças sensoriais entre grupos de amostras, as
variáveis de importância preditiva e as variáveis afetivas importantes no momento da
compra, o que resultou no Artigo I. O Artigo II teve como base o estudo do perfil
físico-químico, que foi realizado através dos métodos de rotina, para quantificar
umidade, cinzas, proteínas e lipídios, e os seguintes métodos instrumentais: HPLC –
para quantificação de lactose, sacarose e glicose; ICP-OES – empregado na
quantificação de sódio, cálcio, fósforo e potássio; CG-MS – utilizado na identificação
de compostos voláteis; e, por fim, colorimetria – apta a quantificar os parâmetros de
qualidade de cor L*, a*, b*, h*, C* e diferença total de cor (ΔE*).A discussão dos
resultados desse segundo artigo empregou o teste de Correlação de Pearson e a
Análise de Componentes Principais (ACP). Os DLs II, V e VI foram os que
apresentaram maior aceitação e características mais semelhantes àquelas do
produto ideal. As Redes Neurais Artificiais demonstraram melhor capacidade
preditiva que o PLSR para identificar as variáveis de importância preditiva, que
foram cor, viscosidade, adesividade oral e sabor caramelo. Foram identificados 32
compostos voláteis, ΔE* demonstrou ser uma variável com boa capacidade de
indicar diferenças do padrão de cor, a sacarose demonstrou ser o glicídio de maior
predominância no produto e foi observado que as quantidades de lactose, cinzas,
lipídios e proteínas estão diretamente relacionadas à presença da matriz láctea.
Através do conhecimento da composição desse produto e da influência que os seus
ingredientes tem sobre o perfil sensorial, o direcionamento das melhorias do
processo produtivo se torna mais racionalizado. Como consequência, a satisfação
do consumidor se torna mais acessível e os órgãos de fiscalização e as indústrias
passam a possuir maior base científica para aprimorar os padrões de identidade e
qualidade do DL pastoso.
Palavras-chave: Doce de Leite, Aceitação, Escala do Ideal, ADQ, Redes Neurais
Artificiais, HPLC, CG-MS, ICP-OES
ABSTRACT
The Dulce de Leche (DL) consists of concentrated milk with sucrose, typically
produced in South America. The objective of this work was to investigate the
sensory, physical and chemical characteristics of pasty Dulce de Leche (DL). From
seven traditional brands in the market, obtained in their original packaging, sensory
acceptance tests were performed using the hedonic scale, the Just-About-Right
scale (JAR) and purchase intent. The samples were also subjected to quantitative
descriptive analysis (QDA) and different statistical models were used to acess the
sensory similarities between groups of samples, variables of importance in projection
(VIP’s) and major affective variables at the time of purchase, which resulted in Article
I. Article II was based on the study of physicochemical profile, which was
accomplished through the routine methods to quantify moisture, ash, protein and fat,
and the following instrumental methods: HPLC - for quantification of lactose, sucrose
and glucose; ICP- OES - used in the measurement of sodium, calcium, phosphorous
and potassium; GC -MS - used to identify volatile compounds; and finally colorimetry
- able to quantify the color quality parameters L*,a*,b* ,h* ,C* and the total color
difference (ΔE*) The discussion of the results of this second test used Pearson
Correlation and Principal Component Analysis (PCA). The DLs II, V and VI were
those with greater acceptance. Artificial Neural Networks have shown best predictive
power than PLSR in identifying variables of predictive importance, which were color,
viscosity, adhesiveness and oral caramel flavor. 32 volatile compounds were
identified. ΔE* proved to be a variable with good capacity in indicating differences in
the pattern of color, sucrose proved to be the most predominant carbohydrate in the
product and it was observed that the amounts of lactose, ashes, lipids and proteins
are directly related to the presence of milk matrix. Through knowledge of the
composition of the product and its ingredients influence that has on the sensory
profile, targeting the improvement of the production process becomes more
streamlined. As a result, consumer’s satisfaction becomes more accessible and
oversight agencies and industries shall possess greater scientific basis to enhance
the standards of identity and quality of the pasty DL .
Keywords: Dulce de Leche, Acceptability, Just-about-right scale, QDA, Artificial
Neural Network, HPLC, CG-MS, ICP-OES.
1 INTRODUÇÃO
O estudo do perfil sensorial e físico-químico é de grande valia para a
indústria laticínia, uma vez conhecendo as características de um produto ideal é
possível se pensar em métodos menos onerosos de produção gerando um produto
altamente competitivo (MACFIE, 2007). Os limites físico-químicos que são seguidos
para a produção do DL são estabelecidos pela atual legislação, em consonância
com a do Mercado Comum do Sul (MERCOSUL), através da Portaria nº354
(BRASIL, 1997), Regulamento Técnico de Identidade e Qualidade do Doce de Leite.
O equilíbrio entre elaborar um produto de alta aceitação, viável
economicamente, e dentro dos padrões físico-químicos estabelecidos pelos órgãos
oficiais é um desafio diário de toda indústria (MACFIE, 2007). Para se ter um maior
controle da qualidade físico-químico, cada vez mais a indústria e os órgãos oficiais
requerem metodologias rápidas e precisas. Junto a essa evolução e à necessidade
que o consumidor vem tendo de saber a composição do produto que ele está
adquirindo, é necessário o estabelecimento de novos parâmetros e limites.
A aceitação do consumidor é crucial para o desenvolvimento,
reprodução e melhoramento de produtos. Tendo isso em vista, as escalas hedônica,
do ideal e de atitude são as análises afetivas mais comumente utilizadas (MACFIE,
2007). Juntamente com a Análise Descritiva Quantitativa (ADQ), que promove a
descrição qualitativa e quantitativa do produto através de uma equipe de julgadores
treinados, os resultados desses testes afetivos fornecem uma base de dados para
se determinar os atributos sensoriais mais importantes na aceitabilidade do produto
pelo consumidor final (STONE; BLEIBAUM; THOMAS, 2012).
Além da análise de variância (ANOVA) procedida de testes de diferença de
média, a estatística multivariada consta como uma importante ferramenta para se
elaborar mapas descritivos e correlações entre diferentes metodologias. A Análise
de Componentes Principais (ACP) vem sendo classicamente utilizada na
representação gráfica da ADQ a fim de destacar as características mais marcantes
respectivas de cada amostra testada (STONE; BLEIBAUM; THOMAS, 2012). Novas
análises multivariadas vêm sendo adaptadas à análise sensorial com o objetivo de
identificar os “drivers” que estão mais relacionados ao comportamento afetivo do
consumidor, obtido através de testes como de aceitação e intenção de compra ou
consumo. Uma ferramenta empregada com sucesso para esta finalidade são os
testes de regressão, sendo de alta adequabilidade à ciência sensorial a Regressão
de Quadrados Mínimos Parciais (PLSR – “Partial Least Squares Regression"),
Logística (LR –"Logistic Regression"), por Componentes Principais (PCR – “Principal
Componente Regression”) e Multilinear (MLR - "Multilinear Regression") (NAES;
BROCKHOFF; TOMIC, 2010). Visando uma melhor adequação aos órgãos dos
sentidos e à complexidade de informações que estes processam, métodos não
lineares vem sendo desenvolvidos. As Redes Neurais Artificiais (ANN - "Artificial
Neural Network") utilizam a tecnologia da informação para simular as diversas vias
nervosas evocadas através das percepções sensoriais com o objetivo de, como nas
regressões, identificar os “drivers” mais importantes para a percepção afetiva do
consumidor. As ANN se moldam com destreza às matrizes alimentares mais
complexas do ponto de vista sensorial, que são altamente heterogêneas entre si,
trazendo resultados verossímeis (CRUZ, 2011; BAHRAMPARVAR; SALEHI;
RAZAVI, 2013).
Em função de haver uma lacuna na literatura recente com relação ao perfil
sensorial e físico-químico do DL, a presente dissertação visa: (i) estabelecer os
“drivers” de aceitação e compra do DL pastoso; (ii) caracterizar sensorialmente o DL
ideal; (iii) testar metodologias estatísticas multivariadas sobre os dados sensoriais
dessa matriz; (iv) caracterizar esse produto físico-quimicamente; (v) testar
metodologias instrumentais para a análise de carboidratos, voláteis e minerais
(respectivamente, HPLC, CG-MS e ICP-OES); (vi) e discutir a influência desses
parâmetros sobre os aspectos sensoriais.
15
2 FUNDAMENTAÇÃO TEÓRICA
2.1 DOCE DE LEITE PASTOSO
O Doce de Leite (DL) é um derivado lácteo produzido basicamente a partir
da desidratação do leite fluido e/ou reconstituído, em condições de temperatura e
pressão que variam de acordo com o fabricante, e adicionado de sacarose. A esse
produto também pode ser adicionado o bicarbonato de sódio. Esse composto
incrementa a coloração amarronzada, típica de compostos gerados a partir reação
de Maillard - que é favorecida quando o pH do DL se encontra entre 6.0 e 7.0. Além
disso, essa adição evita a formação de grumos resultantes da desestabilização da
caseína, dada quando compostos ácidos são concentrados durante o processo de
evaporação (PERRONE, 2011). A reação de caramelização também interfere na cor
do DL devido ao seu alto teor de glicídos.
O DL é fabricado basicamente na América Latina, principalmente na
Argentina e Uruguai e, em menor escala, no Chile, Paraguai, Brasil e Bolivia
(ZALAZAR; PEROTTI, 2011). Além do consumo interno, o Brasil exporta para outros
mercados, como para a União Européia (EU) e para a Cooperação Econômica da
Ásia e do Pacífico (APEC). A sua importância nutricional está relacionada à
concentração dos diversos componentes do leite, como proteínas, glicídios e sais
minerais, além da sacarose que, juntamente à lactose, oferece ao DL a propriedade
de ser um alimento altamente energético (OLIVEIRA, 2009).
2.2 ANÁLISES FÍSICO-QUÍMICAS
No Brasil alguns autores já avaliaram os aspectos físicos e químicos do DL,
como DEMIATE, KONKEL e PEDROSO (2001) e SILVA et al.(2007), encontrando
uma enorme variedade de padrões dessa matriz.
Os requisitos físico-químicos básicos estabelecidos pela Portaria Nº 354
(BRASIL, 1997) são: umidade (no máximo 30% - 30g/100g de doce de leite); matéria
gorda (entre 6,0 e 9,0% - 6,0 a 9,0g/100g de doce de leite); cinzas (no máximo 2,0%
16
- 2,0g/100g de doce de leite); e proteínas (no mínimo 5,0% - 5,0g/100g de doce de
leite).
A cor dos derivados lácteos é determinada por diversos fatores variando
desde os relacionados à alimentação do rebanho (composição da matéria-prima) até
a adição de corantes artificiais feita pelas indústrias (WADHWANI; McMAHON,
2012). Esse é um atributo sensorial de extrema importância, principalmente naquilo
que tange à aparência, influenciando diretamente na aceitação por parte do
consumidor (STONE; BLEIBAUM; THOMAS, 2012). Generalizando, a cor e a
aparência do alimento criam uma expectativa no consumidor que afeta o modo como
ele observa e se comporta diante do produto (WADHWANI; McMAHON, 2012),
também exercendo influência direta nos padrões de identidade e qualidade dos
derivados (DELWICHE, 2004).
Glicídios são compostos orgânicos de composição básica de carbono,
hidrogênio e oxigênio, mas que também pode possuir outros elementos associados,
e estão vinculados a dietas energéticas (FEENEMA, 1996). Os glicídios simples são
chamados açúcares. Estes, como glicose, sacarose e lactose, possuem
características que podem influenciar diretamente na qualidade sensorial do produto.
Dentre estas características, se destacam as seguintes: poder edulcorante,
solubilidade em água e fácil formação de xaropes, formação de cristais quando a
água é evaporada a partir de suas soluções, capacidade de serem prontamente
fermentados por microrganismos e de evitar o seu desenvolvimento em altas
concentrações, escurecimento do produto através da caramelização ou Reação de
Maillard ao aquecimento e formação de corpo e paladar diferenciado quando em
solução (POTTER e HOTCHKISS, 1995).
Apesar de existirem alguns estudos acerca da composição centesimal do
DL, existe uma lacuna de informação sobre o seu perfil químico, principalmente
através do uso de técnicas instrumentais como a espectrofotometria de massas
acoplada à Cromatografia Gasosa (CG-MS), a espectrometria de emissão óptica
com plasma indutivamente acoplado (ICP-OES) e a cromatografia líquida de alta
eficiência (HPLC) (OLIVEIRA, 2009). A CG-MS possui valor na identificação de
componentes voláteis dos alimentos ao se obter o índice de retenção linear (LRI)
dos compostos retidos na coluna (CONDURSO, 2008; PANSERI; 2011; PADILLA,
2014). A ICP-OES é uma técnica que que permite a identificação e quantificação de
17
elementos presentes tanto em grandes, quanto pequenas quantidades e até mesmo
traços em matrizes alimentares complexas (NAOZUKA, 2011). Dentre os
componentes do DL, os carboidratos possuem maior importância devido à influência
que o processamento tecnológico exerce sobre eles. Os dois carboidratos mais
importantes do DL são a lactose e a sacarose (DEMIATE, 2001; OLIVEIRA, 2009).
No entanto não são encontrados na literatura recente trabalhos com DL
quantificando carboidratos por cromatografia líquida de alta performance (HPLC),
que é comumente vista em outros derivados lácteos, como leite (GIMÉNEZ, 2008;
ERICH, 2012) e iogurte (CRUZ, 2013; VÉNICA, 2013; CRUZ, 2012).
18
2.3 ANÁLISES SENSORIAIS
A análise sensorial é uma ciência que possui aplicabilidade em "marketing",
uma vez que permite correlacionar padrões de tecnologia de fabricação com as
impressões que o produto causa no consumidor (MACFIE, 2007; STONE;
BLEIBAUM; THOMAS, 2012). Os testes sensoriais afetivos, com consumidores,
buscam quantificar os impactos positivos e negativos que os produtos teste geram.
Dessa forma, atuam como ferramentas importantes na obtenção do perfil sensorial
de um produto adequado ao mercado consumidor. Para tal, podem ser feitas
correlações entre a aceitação e a escala do ideal (“Just-about-right scale” - JAR),
através da Análise de Penalidades (PA) (CADOT, 2010; LEE, 2011). O método
estatístico de Regressão Logística (LR) pode ser utilizado para correlacionar os
diversos testes afetivos à escala de intenção de compra (como uma variável
resposta dicotômica), fornecendo informações sobre quais atributos são mais
importantes no momento da compra (GARCIA, 2009; CRUZ, 2011). Os testes
sensoriais descritivos podem ser utilizados para se obter as intensidades de
percepção de descritores, que são classicamente levantados através da Análise
Descritiva Quantitativa (ADQ), empregando um painel treinado (STONE; BLEIBAUM;
THOMAS, 2012). Com a finalidade de rebuscar o perfil sensorial do produto, os
testes afetivos e os descritivos podem ser correlacionados entre si através de
estatística multivariada, como o Mapa de Preferência (MACFIE, 2007), Regressão
por Mínimos Quadrados Parciais (PLSR) (KRISHNAMURTHY, 2007; BAYARRI,
2011; VILANOVA, 2012) e Redes Neurais Artificiais (ANN) (KRISHNAMURTHY,
2007; BAHRAMPARVAR, 2013). Dessa forma, se torna possível chegar a
conclusões sobre quais atributos e suas respectivas intensidades têm maior
influência na aceitação do produto.
19
3 DESENVOLVIMENTO
3.1 ARTIGO 1: OPTIMIZATION OF THE DULCE DE LECHE PROCESSING: CONTRIBUTIONS OF THE SENSOMETRIC TECHNIQUES
20
Optimization of the dulce de leche processing: contributions of the Sensometric techniques
L.V. Gazea*, B.R. Oliveiraa, D. Granatob, C.A.Conte Júniora, A.G Cruzc , M.Q. Freitasa
a Universidade Federal Fluminense (UFF), Faculdade de Veterinária, Pós-Graduação em Higiene Veterinária e Processamento de Produtos de Origem Animal, 24230-340 Niterói, Rio de Janeiro, Brazil
b Universidade Estadual de Ponta Grossa (UEPG), Departamento de Engenharia de Alimentos, Av. Carlos Cavalcanti, 4748, 84030-900, Ponta Grossa, Brazil.
c Instituto Federal de Educação, Ciência e Tecnologia do Rio de Janeiro (IFRJ), Equipe de Alimentos, Rua Senador Furtado, 171/175, Maracanã, 20270-021 Rio de Janeiro, Brazil
*Corresponding author: L.V. Gaze Rua Vital Brazil Filho, nº 64 Santa Rosa - Niterói, Rio de Janeiro, Brasil – Zip code: 24.230-340 Phone.: +55 - 21 - 2629-9545 E-mail: [email protected] (L.V. Gaze)
21
ABSTRACT
Sensometric techniques for the production of dulce de leche with optimal
sensory profile and high acceptance by consumers were evaluated. Dulce de leche
samples available in the Brazilian market were submitted to the development of
sensory profile by quantitative descriptive analysis and acceptance test, as well
sensory evaluation using the just about right scale (JAR) and purchase intent.
Moreover, the ideal sensory characteristics of dulce de leche were determined. The
results were evaluated by principal component analysis (PCA), hierarchical cluster
analysis (HCA), partial least squares regression (PLS), artificial neural networks
(ANN) and logistic regression (LR). Overall, a significant product acceptance is
related to intermediate scores of the sensory attributes in descriptive test, and this
trend was observed even after consumer’s segmentation. However, there is need for
standardization of certain attributes, since no product was characterized as ideal for
the attributes evaluated in this study. The results obtained by sensometric techniques
showed that optimizing an ideal dulce de leche from the sensory standpoint is a
multidimensional process, with necessary adjustments on the appearance, aroma,
taste and texture of the product for better consumer’s acceptance and purchase. In
industrial terms, this means changing the parameters used in the thermal treatment
and quantitative change of the ingredients used in the formulations.
Key-words: sensometric technique, dulce de leche, optmization
22
INTRODUCTION
Multivariate statistical techniques are very useful for analyzing complex data
obtained for consumer and trained panel, as they are able to provide a spatial
representation of the experimental data (Ennis and Ennis, 2013, Zielinski et al.,
2014). In addition, they proportionate the acquisition of the drivers of consumer liking
and purchase intent, which is a popular concept frequently used in sensory and
consumer research (Bi, 2013). Indeed, when it is noted differences among foods
products that may drive consumer liking and purchase intent, it is helpful to visualize
a space in which each possible product is placed (Cadena et al., 2012). Another
outcome is consumer's food choices, which are complex interactions between the
product and the consumer as well as the consumer’s personality and the previous
familiarity and experience with the product (Cadena et al., 2013)
Dulce de leche (DL) is a dairy-based product, produced mainly from the
dehydration of fluid milk under conditions of temperature and pressure varying
according to the manufacturer, with added sucrose. Sodium bicarbonate can also be
added to the milk, increasing the typical brown color caused by the Maillard reaction,
which is favored at pH 6-7, and avoiding the formation of lumps resulting from the
concentration of acidic compounds generated during evaporation (Perrone et al.,
2011). DL is manufactured and consumed in Latin America, especially Argentina and
Uruguay and to a lesser extent in Chile, Paraguay, Brazil and Bolivia (Zalazar and
Perotti, 2011). In Brazil, the production is performed exclusively by small dairy
producers, without control of the operational parameters, resulting in products with
different physicochemical properties, with direct influence on the sensory
23
characteristics. The consumer market is basically formed by two groups: the end
consumers and the institutional consumers, formed by the food service and the food
industry. The institutional segment is a large market, represented by fast food chains,
ready prepared food, ice cream shops, confectionery, and wholesale centers. In
addition, hotels, motels, convenience stores and board services have also acquired
this product in the form of mini portions (Milkpoint, 2013).
The product quality depends on the continuous consumer satisfaction with
respect to the sensory attributes, thus it is necessary to evaluate the relevant
attributes for acceptance and purchase intent of a food product. In this context, this
study aims to identify the possibilities of improvement of dulce de leche available in
the Brazilian market, identifying relevant sensory attributes for acceptance and
purchase intent. Given the multidimensional nature of the study, multivariate
statistical techniques were used to analyze the results, such as Partial Least Squares
Regression (PLS), artificial neural networks (ANN) and logistic regression (LR). The
results of this study may be useful to provide information for dairy companies to make
decisions, such as the development and marketing of new products and / or the
reformulation of existing products, besides establishing quality control specifications.
MATERIAL AND METHODS
Samples
Seven commercial brands of dulce de leche were purchased from dairy
companies inspected by the Brazilian Federal Inspection Service (SIF), which
enables the marketing throughout the Brazilian territory. The samples were acquired
right after processing so shelf life was not considered as a factor in this study. Then,
24
samples were coded as I, II, III, IV, V, VI and VII to maintain the confidentiality of the
trademarks. The ingredients used in manufacturing the product, as well as the
nutritional composition displayed on the label, are shown in Table 1. The samples
were presented in randomized and balanced manner for both the descriptive analysis
as consumer tests, according to Mac Fie et al. (1989).
Sensory profiling (Quantitative Descriptive Analysis, QDA)
To quantify the descriptive attributes of the samples, Quantitative Descriptive
Analysis (QDA) was performed. This technique is traditionally used in the evaluation
of sensory profiling of processed foods, especially dairy products (Kaaki et al., 2012;
Albenzio et al., 2013; Murtaza et al., 2013; Pimentel et al., 2013).
In the recruitment stage, 40 assessors were evaluated regarding their oral
health status, interests, schedule availability, ability to verbal descriptions of sensory
perceptions and use of scales. In the selection stage, a triangle test was used to
check for differences (Stone et al., 2012), in which assessors evaluated the
perception of sweetness in water solutions containing different concentrations of
sucrose (0.45% and 0.65% w/v). Four replications were carried out, in which the
assessors should present at least 75% accuracy. Fifteen assessors were selected for
the training stage. During training, the sensory attributes were collected using the
Kelly's repertory grid method (Morais et al., 2014a). Then, a meeting with the
assessors was performed to define the final list of attributes as well as their
references of weak and strong intensity. The training stage consisted of nine
sessions lasting an hour each, totaling 9 hours. A descriptive vocabulary was drawn
up and the intensity terms and reference materials were defined (Table 2)
(Moskowitz, 1983). The performance was monitored at the end of the training
25
sessions in triplicate using the samples with markedly different sensory
characteristics and presenting the pre-established attributes. Data were obtained
through a non-structured evaluation sheet containing the attributes in a 15 cm linear
scale. From these data, 5 assessors were eliminated, forming a trained team of 10
assessors, 4 men and 6 women aged between 23 and 42 years, which meets the
recommendations of Heymann et al (2012), who suggested at least 8 and preferably
10 assessors as a suitable number for implementation of QDA. The whole process
was conducted in individual booths at 25 °C, with the samples presented in
randomized and balanced way in disposable plastic cups coded with three-digit
numbers, for trial of oral attributes (Mac Fie et al, 1989.); in watch-glasses for
appearance attributes; and in beakers covered with watch-glasses for aroma
attributes. Unsalted cracker biscuits (Piraquê, Rio de Janeiro, Brazil) and water for
cleansing the palate were provided at all stages.
Consumer test
The consumer tests were carried out under laboratory conditions, at the
Sensory Analysis Laboratory of the Universidade Federal Fluminense (UFF), with
samples served in plastic cups in a randomized and balanced manner. Cracker
biscuits and mineral water were provided for cleansing the palate between the
samples. One hundred and twenty-five consumers participated in the trial, aged
between 15 and 57 years, being 37 men and 88 women.
Before the sensory tests, the assessors completed and signed a written
consent to participate the tests, stating that they were physically fit for that.
Acceptance testing was conducted using a nine-point hedonic scale (9 - like
extremely, 1 - dislike extremely) to assess the following attributes: appearance,
aroma, taste, texture and overall acceptance, and the purchase intention was
26
assessed by a five-point scale (1- definitely would not buy, 5 - I definitely would buy)
(Cruz et al, 2012). The attributes color, sweet taste, caramel taste, and consistency
were evaluated using just about right scale of 9 points, with 9 - extremely over than
the ideal; 5 - Ideal, and 1 - extremely less than ideal (Morais et al., 2014b; Chollet et
al., 2013).
Data analysis
Overview of sensometric techniques
Partial Least Squares Regression (PLS) is a method widely used in sensory
analysis using both consumer and sensory descriptive analysis by a trained panel,
enabling the identification of positive and negative attributes that contribute to
product acceptance (Cadena et al., 2012). PLS components are created by the
explanatory variable (X-data) and dependent variable (Y-data), being the original
data decomposed into principal components called factors (Cruz et al., 2011). These
factors are linear combinations of the original predictive variables and are calculated
so that a maximum amount of variation in the x variables (sensory attributes) is
explained (Meullenet et al., 2007). However, it presents linear character, and does
not adequately address non-linearity, which is common in the analysis of QDA and
consumer liking data (Cruz et al., 2011).
Artificial neural network (ANN) is a mathematical algorithm capable of
relating input and output parameters through training, without requiring a prior
knowledge on the relationships between the process variables (Cevoli et al., 2011).
ANNs consist of a number of simple processing units (or neurons) interconnected by
a modifiable weight, originally designed to mimic the function of the human brain and
applied to quantitative and qualitative analysis during the last decades. ANN requires
a training process in which the weights of those connections are adjusted, building a
27
model that will allow performing the prediction of the desired parameters (either
qualitative or quantitative). Its main advantages include a high modeling
performance, being especially suited for nonlinear systems, besides being very much
related to human pattern recognition (Cetó et al., 2013). Between the input and the
output layers, there is at least one hidden layer with any number of neurons. The
number of neurons in the hidden layer (s) depends on the application of the network.
In the hidden and output layers, the net input (xj) to node j is of the form:
j
n
1i
iijj byWfX
Where yi are the inputs, W ij are the weights associated with each input/node
connection, n is the number of nodes, and bj is the bias associated with node j.
Additionally, bias is an extra input added to neurons. The reason for adding the bias
term is that it allows a representation of phenomena having thresholds. Each neuron
consists of a transfer function expressing internal activation level. Output from a
neuron is determined by transforming its input using a suitable transfer function. The
transfer function can be linear or nonlinear (commonly sigmoid and hyperbolic
tangent) functions, depending on the network topology (Bahramparvar et al., 2013).
Penalty (mean-drop) analysis is used by market researchers and product
developers to understand the product attributes that most affect liking, purchase
intent or any other product-related measure, being obtained from the Just about right
data. Penalty analysis (PA) measures the changes in product liking (or any other
measure) due to the product having “too much” or “too little” of a specific attribute
(Market Research World, 2013). It has been used extensively by practitioners in the
industry to identify the lower acceptance of a product associated with sensory
attributes not at optimal levels (Sensory Society, 2013). There are few reports about
28
using penalty analysis in dairy foods consumer study, being related twos studies
covering dairy foods: cottage cheese (Drake et al., 2011) and vanilla yogurt added
with stevia (Narayanan et al., 2014).
Logistic regression (LR) is a powerful tool for the statistical analysis of
categorical data with the main purpose to discover exploratory variables that drive a
choice outcome, without necessity of evaluating integrals (Ennis and Ennis, 2013). It
can be used to correlate affective tests, in general, the purchase intention (which in
this case is presented as a dichotomous response variable), providing information
about the most important attributes at time of purchase (Prinyawiwatkul and
Chompreeda, 2007). The sensory attributes can be used as limiting factors for
obtaining optimal formulations.
Statistical analysis
The consumer test and the QDA means values were analyzed by one way
analysis of variances (ANOVA) and Tukey’s mean difference (P < 0.05) test,
considering the sample as fixed effect (Granato et al., 2014). The QDA data were
also subjected to principal components analysis (PCA) using the correlation matrix,
while the hierarchical cluster analysis (HCA) was applied to the overall liking data to
identify groups of consumers with similar preferences. In HCA, Euclidean distances
(dissimilarity), Ward's method (agglomeration method) and automatic truncation were
employed (Cruz et al., 2013).
To identify the attributes that were related positively with the product
acceptance, two different methodologies were used: partial least squares regression
(PLS) and artificial neural networks (ANN) considering that both methods have
opposite behavior with respect to linearity of the data. In both models, the overall
liking was coded on binomial scale (0/1), representing respectively the rejection,
29
value 0, overall liking values among 1.00 to 5.99, and acceptance, value 1, overall
liking values among 6.00 to 9.00, Vidigal et al., 2011). To identify natural variation
among consumers in the product acceptance test, the variables were not subject to
any pre-processing of data. The PLS model quality was evaluated by three
parameters: coefficient of determination (R2), related to the variation in the dependent
variable explained by the regression model and the cumulative index (Q2), which
measures the global contribution of the ‘h’ first components as to predictive quality of
the model (Donadini et al., 2012; Aquino et al., 2014).
ANN architectures and topologies were assayed employing Bayesian
regularization algorithms, which used different number of neurons (5-55 neurons)
and training algorithms (Scaled Conjugate Gradient and Levenberg - Marquardt) in
the hidden layer, and different activation functions in the hidden and output layer
(tangent, identity, logistic, softmax, sigmoid and exponential). The number of input
neurons corresponds to the number of input variables into the neural network, and
the number of output neurons is similar to the number of desired output variables.
The input dataset was comprised by the sensory attributes obtained from QDA. The
output layer included two neurons corresponding to two possible acceptance classes,
which can be classified according to the overall liking (OL) as accepted (OLG > 6.0)
or rejected (OLG ≤ 6.0 ). The target output was 1.0 in the correct class output, and
0.0 in the other. The outputs are probabilities of membership in every class, provided
that the total across all output units is 1.0. The original datasets (280 samples) were
randomly divided into training set (75%) and test set (25%) for all network topologies
tested. The training set was used to calculate the transfer function parameters of the
network, and the test set was used to estimate the correct classification in which the
neural network is performing well. An ideal network would classify correctly all sets.
30
The number of neurons in the hidden layer was obtained by trial and error. The
stopping criterion was 1000 epochs or maximum sum of square errors (SSE), Eq. (1),
equal to 0.001 during training. Various performance measures are computed during
the test. Each predicted value was compared with the experimental value to test the
network performance. For this purpose, mean square error (MSE) was calculated
using Eq. (2). On the lower values of MSE, the network predicts the values more truly
and the correlation coefficient (r), Eq. (3), gives information about the performance. If
the correlation coefficient is close to one, it shows how much the learning and
prediction is successful.
N
1i
2
ii TOSSE
(1)
N
TO
MSE
N
1i
2
ii
(2)
N
1i
2
mi
N
1i
2
ii
TO
TO
1r
(3)
Where Oi is the ith actual value, Ti is the ith predicted value, N is the number
of data, and Tm is given by:
N
O
T
N
1i
i
m
(3)
31
A confusion matrix was performed to determine sensitivity (true positive rate),
specificity (true negative rate) and accuracy values in order to evaluate the
classification performance of the binary classifier system (Xu et al., 2012). The
sensitivity (Sens), specificity (Spec) and accuracy (Accu) were determined as follow:
Sensitivity, specificity and accuracy analysis were used to evaluate the
performance of the pattern classification techniques for each class (Xu et al., 2012).
Sensitivity (Sens), specificity (Spec) and accuracy (Accu) were determined as
follows:
FNTP
TPSens
…...(4)
FPTN
TNSpec
(5)
FPFNTPTN
TPTNAccu
(6)
Where TP, FN, TN, and FP denote the numbers of true positives, false
negatives, true negatives, and false positives, respectively. For calculations purpose,
samples classified as accepted were considered as “positive”, and the rejected
samples were considered as “negative”. Furthermore, a sensitivity analysis was
performed to provide a measure of the relative importance among the inputs of the
neural network model and to illustrate the model output variability as a function of the
change in the input variable.
JAR scores were evaluated through the penalty analysis, being considered
significant when more than 20% consumers evaluated the sample above or below
32
the Just Right (Drake et al., 2011; Narayanan et al., 2014). The purchase intent data,
computed in binomial form (1 representing buy, > 3 and 0, representing not buy, ≤ 3),
were correlated with data from the JAR scale and acceptance test by Logistic
Regression (LR), in order to find the affective characteristics that most influenced the
purchase intent of the product (Cruz et al., 2011).
Except for neural networks, all analyses were performed using the software
XLSTAT version 2013 (Addinsoft, Paris, France). The artificial neural networks were
modeled by the software Statistica v. 8.0 (StatSoft Inc., Tulsa, OK, USA) using a
supervised multilayer perceptron (MLP) network to predict the sensory acceptability
of DL samples obtained using QDA (Cruz et al., 2011; Cruz et al., 2009).
RESULTS AND DISCUSSION
Sensory profile
Table 3 shows the average attributes obtained in the quantitative
descriptive analysis of dulce de leche. The sensory profile was composed of fifteen
descriptors, namely apparent adhesiveness, apparent viscosity, color, brightness,
heated milk aroma, caramel aroma, sweet taste, heated milk taste, caramel taste,
butter, pungent aftertaste, oil layer, adhesiveness, viscosity and sandiness.
Significant differences were observed between all attributes for all samples (P <
0.05). In particular, significant difference was also observed for some sensory
descriptors such as brightness (ranging from 2.26 to 13.07), viscosity (ranging from
0.865 to 13.32), and sandiness (ranging from 0.705 to 13.82) among others. With
respect to the samples, the DL samples I, IV, and VII showed extreme scores on the
attributes brightness, color, adhesiveness and viscosity (apparent and oral) and
caramel and heated milk (aroma and taste) while DL II presented significant scores
for the attribute sandiness. Finally, the samples V and VI showed a higher number of
33
attributes with intermediate scores. Overall, the results suggest that there is a wide
variety of products offered to the consumer resulting from different formulations and
parameters used in the manufacturing process, such as intensity of heat treatment
and amount of sodium bicarbonate, sucrose and glucose, in which the latter is a
facultative ingredient.
The two dimensional principal component analysis (PCA) accounted for
87.22% (68.34% and 18.88 % for dimensions I and II, respectively, as shown in
Figure 1), and sample II was characterized by the attribute sandiness, and to a lesser
extent by the attributes apparent adhesiveness, viscosity, and adhesiveness. The
sample IV was characterized by pungent aftertaste, color, caramel taste and aroma,
while samples V and VI presented sweet taste, caramel taste and aroma, and color.
With respect to the samples I and III, they were characterized by buttery taste and oil
layer, while the sample VII was characterized by heated milk aroma and heated milk
taste. As can be seen in Figure 1, the first principal component is characterized by
the attributes brightness, pungent aftertaste, butter taste, and viscosity, while the
second principal component is characterized by sandiness, heated milk aroma and
color.
Consumer test
The results of the consumer’s test are shown in Table 4. Corroborating the
data from the sensory profile obtained by QDA, significant differences were observed
for all attributes (P < 0.05), suggesting again heterogeneous products from the
sensory point of view. The sample VI followed by the samples V and II showed the
highest values for the overall acceptance (7.5, 6.8, and 6.5, respectively, P < 0.05),
whose behavior was also observed for appearance (7.5, 6.6, and 7.2, P < 0.05,
respectively), aroma (6.8, 6.6, and 6.4, P < .05, respectively), taste (7.4, 7.0 and 6.3,
34
respectively, P < 0.05) and texture (7.6, 6.9, and 6.2, P < 0.05, respectively). In
contrast, the samples I and VII presented lower scores for overall appearance (4.6
and 4.1, respectively, P < 0.05), ranging from 4 (extremely disliked) to 6 (liked
slightly) for the other attributes, with no attribute scored above 6, which represents
the lower limit of acceptance zone on a 9-point scale (Milagres et al., 2011).
Interestingly, the samples II and VI showed higher scores for appearance, as
opposed to the sample VII, which had the lightest color according to the trained
panel, thus evidencing the importance of color in the product appearance. The higher
scores found for the sample VI in overall impression may be related to their
descriptive characteristics that, in general, did not present discrepant mean values,
given the expectation of a greater number of consumers. It is important to emphasize
that the composition of sample VI declared on the label is quite simple, once besides
milk and sucrose, the formulation contained only sodium bicarbonate and potassium
sorbate. This fact indicates the possibility of manufacturing a dulce de leche with
better acceptance and simplified composition, facilitating the widespreading of the
product and reducing manufacturing costs by using a less complex and more
economical technology. It may also be noted that despite it is not within the values
recommended for dulce de leche, the attribute sandiness did not have relevant
influence on the overall impression, since the DL II was accepted and showed
significant scores for this attribute.
Cluster analysis
Cluster analysis was performed with the aim of targeting based on their
global acceptability consumers. Three clusters CL1, 2 and 3 were identified with 74,
24 and 47 consumers, corresponding to 59.2, 19.2 and 37.6% respectively (Table 5).
35
Since 88 of the 125 consumers were women (70.4%) in all clusters was observed
predominance of womem. With respect to age, due to the fact the study was
developed in a university center, was also observed in general predominance of a
younger audience, aged 18-24 (68.5 and 59.6% of consumers present in clusters
CL1 and 3 respectively while for cluster II predominance was observed for an
audience between 25-34 years (corresponding to 58.3% of consumers in this cluster
members). Regardless of the segment formed, samples II, V and VI showed the
highest scores for overall acceptability (values between 7.1 and 8.1, respectively).
Analyzing the differences between clusters for each food (P <0.05), one can observe
that the cluster I have a greater tendency to accept the DLs regardless of how they
are heterogeneous among themselves. The cluster I gave the highest scores of
acceptance for all seven DLs, with statistical difference (P <0.05), compared the
other two clusters, except CL2 in the sample V and CL 3 in sample II. The relevance
of CL 1 is its high concentration of consumers. Additionally, the high score of the
sample "II" in the cluster III, suggests that sandiness is not a relevant parameter for
rejection of that product. This reinforces the findings of previous studies where it was
observed that sandiness is tolerated in different ways with respect to consumers of
fresh milk (Giménez et al., 2008).
JAR scale and penalty analysis
Table 4 shows the penalty analysis performed using data from the just-about
right-test for the attributes color, sweet taste, caramel taste, and consistency. The DL
VI showed smaller mean drops for all the attributes evaluated (0.4, 0.5, 0.7, and 0.8,
respectively), confirming its high acceptability demonstrated by the overall
impression, and suggesting an optimized formulation from the sensory point of view
36
of the consumers. The DL VII showed the highest response percentages in color
below JAR (94.40%), and greater mean drop value used to estimate the penalty due
to its below JAR (3.1), which were the highest among all the samples. This
corroborates their negative acceptance in the overall impression and shows that the
attribute color is of utmost importance as a criterion for sensory quality in the eyes of
the consumer. Opposite behavior produced the same effect, as the excessive
browning of DL IV showed high frequency (89.6%) and mean drop (2.3), considering
it much too brown, above JAR. Probably, since it is a traditional product containing
high carbohydrate levels with sweetening power, all DLs were considered sweet
taste above JAR, with a frequency response above 20% of the population. Although
lower frequencies were reported, the characteristic sweet taste below JAR had larger
mean drops when compared to too much sweet taste, indicating that the loss of
expectation with respect to this attribute had a greater influence on the product
acceptance. The higher frequency of caramel taste below JAR was found for the
sample VII (82.4%) that, as occurred earlier in JAR color attribute, presented a high
mean drop (2.3), indicating the strong influence of this attribute on its low
acceptance, and a direct relationship with the color due to excessive Mailard
reaction. The DL VI showed higher response percentages of caramel taste JAR
(61.6%) while the attribute consistency most influenced the DL IV overall liking, with
greater frequency (81.6%) and greater mean drop (2.4) for consistency above JAR
when compared to the previous attributes. The DL I was considered less consistent,
obtaining greater mean drop for such attribute (1.7).
In general, DLs I, IV, and VII presented the lowest mean values for
overall liking. The mean drops indicate that this may be due to the answers sweet
taste below JAR and caramel taste below JAR for the DL I, brownish above JAR and
37
consistency above JAR for the DL IV, and brownish below JAR and caramel taste
below JAR for DL VII.
According to Meullenet et al. (2007), a specific attribute is present in
optimal levels in a product when a minimum of 70% of the answers are in the "JAR"
group. Based on this, the response frequencies show that the DL VI can be
considered in JAR category with respect to its consistency, since it reached 74.40%
of the responses, although a significant number of responses was also observed for
the attribute caramel taste, with 61.60%. Corroborating the above, the results
indicate the need for improvements in the attributes for all samples, indicating the
imminent need for standardization of the manufacturing stages of product.
Drivers of liking using Partial Least Squares Regression (PLS) and
Artificial Neural Networks (ANN)
The use of PLS generated three components, with Q ² acum, R ² Xacum, and
R ² Yacum values of 0.564, 0.719 and 0.411, respectively. Thus, it indicates low
predictive ability of the model and a little explanation of the variability of the data,
which may be related to the formulation heterogeneity, which contributed to the
nonlinear behavior data, once samples with high scores in certain descriptors
presented opposite behavior for other descriptors. In this context, the PLS does not
seem adequate to estimate the drivers of liking of dulce de leche, therefore a non-
linear method was needed, such as artificial neural networks.
The ANN performance was evaluated using prediction accuracy and
mean square error (MSE) of training and testing sets. The proximity of the accuracy
values to 100% and closeness of the mean square error values to 0 shows the
efficiency of the prediction of ANN model. The best ANN architecture was composed
38
by 15 input neurons, 5 hidden neurons and two output neurons using logistic and
hyperbolic tangent function in hidden and output layer, respectively. The receiver
operating characteristic (ROC) curve is a graphical plot, which illustrates the
performance of a binary classifier system as its discrimination threshold is varied.
The ROC curve analysis achieved sensitivity (true positive rate), specificity (true
negative rate) and accuracy values of 98.0, 100.0, and 98.6% for all dataset. The
high accuracy of the ANN model can be corroborated by the low MSE value of
1.73%, with 100% of recognition and prediction ability for the accepted class, while
the rejected class presented 95% and 93% of recognition and prediction ability for
training and testing dataset, respectively. The sensitivity analysis showed color,
apparent viscosity, adhesiveness, and caramel taste as the sensory attributes with
the most discriminative power, presenting variable of importance in projection (VIP)
coefficients of 7.009, 3.915, 3.501, and 2.321 for all dataset, respectively. In this
context, improving the acceptance of Dulce de leche requires a reformulation of the
ingredients and process parameters that influence these attributes, such as control of
temperature and time of heat treatment and addition of ingredients such as sucrose,
glucose and sodium bicarbonate.
External preference mapping
Figure 2 shows the external preference mapping obtained from the
regression of the acceptance data of each cluster on the 2-dimensional principal
component analysis. For cluster 1, a circular model was obtained by the existence of
an ideal point (-0.723, 0.542), while quadratic and elliptical models were obtained for
CL2 and CL3, with the existence of saddle points that are shown on the map. For a
specific individual, the ideal point shows the location of his/her most preferred sample
39
(a global maximum in terms of preference), while the saddle point represents a
threshold where the variability of the preference is low, before increasing or
decreasing in the opposite direction (Resano et al., 2010).
As shown in the two-dimensional map, samples V and VI are located in the
region that comprises acceptance values above the overall value for 80-100% of
consumers. In this region, there is the presence of the ideal point, and these samples
are associated with the attributes color, sweet taste, caramel taste, and caramel
aroma. The saddle point region precedes the region in which the acceptance values
were 60-80 % above the overall acceptance, where the samples II and III are placed.
The sample II associated with sandiness to a greater extent, followed by the
attributes apparent and oral adhesiveness, apparent and oral viscosity while the
sample III is related to oil layer and butter taste.
Sensory profiling of the ideal products
Besides establishing the sensory profile and the characteristics that drive the
products acceptance by consumers, there is a need to determine the sensory
characteristics that should be exhibited by the ideal products. This procedure is
performed by reverse regression, which in our case it was a PLS regression of the
sensory scores on either the preference space coordinates (obtained from circular
model - CL1) of the ideal product in their respective spaces.
It was observed that the optimum Dulce de leche is character
ized by high scores for the attributes sweet taste, caramel taste, brightness,
color, and caramel aroma (10.3, 9.1, 9.0, 8.5, and 8.4, respectively), intermediate
scores for apparent adhesiveness, heated milk taste, apparent viscosity, heated milk
aroma, and adhesiveness (7.1, 6.8, 6.7, 6.6, and 6.0, respectively), and low scores
40
for viscosity, pungent aftertaste, butter taste, oil layer, and sandiness (6.0, 5.8, 4.8,
4.6 and 2.2 respectively), considering a 15 cm scale.
From an industrial perspective, intermediate heat-treatments may be
used, since attributes such as caramel color, aroma and taste are also influenced by
this operational parameter, together with the formulation ingredients, such as
sucrose, glucose (which contributes to the brightness and browning) and sodium
bicarbonate (acidity regulator, which positively influences the Mailard reaction).
Purchase intent using Logistic Regression (LR)
The Table 6 shows the attributes that influenced the purchase intent of Dulce
de leche using logistic regression. One predictive value of 87.54%, equivalent to a
good ability to fit the experimental data was observed. The results suggest that
buying products is a multidimensional parameter, influenced by mainly two attributes
including overall liking and taste, with odds-ratio values of 3.23 and 1.81,
respectively. The higher scores for the attributes overall liking and taste suggest that
these attributes are more critical for the purchase of the product, with a purchase
probability of 3.23 and 1.81 times higher (than not being purchased, P < 0.0001), with
every one-unit increase of the overall liking score (based on a nine-point hedonic
scale). However, the texture is also relevant, once the probability of the product being
purchased increases 1.21 times (than not being purchased, P < 0.034), resulting in
one-unit increase of the overall liking score.
Overall, overall liking is not easy to understanding when it is related the
purchase intent. So, the logistic regression become easy to understand because they
show the importance of taste and texture that are commonly observed in processed
food products. Additionally, in agreement with the results obtained in the penalty
41
analysis, there is a need to reformulate the taste (caramel and sweet) and
consistency of the dulce de leche samples, which were not considered ideal for these
attributes.
42
CONCLUSIONS
The expansion and conquest of new consumer markets requires the
standardization of sensory characteristics of processed products, resulting in greater
consumer’s acceptance.
In the case of dulce de leche of the present study, a large heterogeneity was
observed from the sensory point of view, and the products most appreciated by
consumers showed intermediate scores for the sensory attributes established in the
descriptive test, and this trend was observed even after consumer’s segmentation.
However, there is still a need for an effective optimization, since no product was
considered as ideal for the attributes under study.
In general, the sensometric methodologies proved to be useful, and indicated
that the sensory optimization of the dulce de leche is a multidimensional process,
requiring simultaneous adjustments in the attributes appearance, aroma, taste, and
texture. From a practical standpoint, the parameters of heat treatment and quantity of
ingredients used in the formulation should be standardized.
The present results may be useful to include dulce de leche on the
production line of dairies, with optimized sensory quality and high probability of
acceptance by consumers, thus resulting in increased revenues for the producing
unit.
43
REFERENCES
Albenzio, M., A. Santillo, M. Caroprese, A. Braghieri, A. Sevi and F.
Napolitano. 2013. Composition and sensory profiling of probiotic Scamorza ewe
milk cheese. J. Dairy Sci. 96: 2792-2800
Aquino, L.F.M.C., A.C.O. Silva, M.Q. Freitas, T.L. Felicio, A.G. Cruz and
C.A. Conte-Junior. 2014. Identifying cheese whey an adulterant in milk: Limited
contribution of a sensometric approach. Food Res Int. 62: 233-237.
Bahramparvar, M., F. Salehi and S. M. Razavi. 2013. Predicting Total
Acceptance of Ice Cream Using Artificial Neural Network. J. Food Proc. Pres., in
press, doi .1111/jfpp.12066
Bi, J. 2012. A Review of statistical methods for determination of relative
importance of correlated predictors and identification of drivers of consumer liking.
J. Sens. Studies. 27:87-101.
Cadena, R.S., A.G. Cruz, R.R. Netto, W.F. Castro, J.A.F. Faria, and
H.M.A. Bolini. 2013. Sensory profile and physicochemical characteristics of mango
nectar sweetened with high intensity sweeteners throughout storage time. Food
Res. Int. 54:1670-1679
Cadena, R.S., A.G. Cruz, J.A.F. Faria, and H.M.A. Bolini. 2012. Reduced
fat and sugar vanilla ice creams: Sensory profiling and external preference
mapping. J. Dairy Sci. 95:4842-4850.
Cetó, X., M. Gutiérrez-Capitán, D. Calvo and M. Del Valle. 2013. Beer
classification by means of a potentiometric electronic tongue. Food
Chem.141:2533-2540
44
Cevoli, C., L. Cerretani, A. Gori, M. F. Caboni, T. Gallina Toschi and A.
Fabbri. 2011. Classification of Pecorino cheeses using electronic nose combined
with artificial neural network and comparison with GC-MS analysis of volatile
compounds. Food Chem. 129:1315-1319
Chollet, M., D. Gille, A. Schmid, B. Walther, and P. Piccinali. 2013.
Acceptance of sugar reduction in flavored yogurt. J. Dairy Sci. 96:5501-5511
Cruz, A.G., R.S. Cadena, W.F. Castro, E.A. Esmerino, J.B. Rodrigues, L.
Gaze, J.A.F. Faria, M.Q. Freitas, R. Deliza, H.M.A. Bolini. 2013. Consumer
perception of probiotic yogurt: Performance of check all that apply (CATA),
projective mapping, sorting and intensity scale. Food Res. Int. 54:601-610.
Cruz, A.G., R.S. Cadena, J.A.F Faria, M. A. Bolini, C. Dantas, M.M.C.
Ferreira, R. Deliza. 2012. PARAFAC: Adjustment for modeling consumer study
covering probiotic and conventional yogurt. Food Res. Int. 45:211-215
Cruz, A. G., R. S. Cadena, J. A. F. Faria, C. A. F. Oliveira, R. N.
Cavalcanti, E. Bona and M. A. A. P. Silva. 2011. Consumer acceptability and
purchase intent of probiotic yoghurt with added glucose oxidase using
sensometrics, artificial neural networks and logistic regression. Int. J. Dairy
Technol. 64:549-556
Cruz, A. G., E. H. M. Walter, R. S. Cadena, J. A. F. Faria, H. M. A. Bolini
and A. M. Frattini Fileti. 2009. Monitoring the authenticity of low-fat yogurts by an
artificial neural network. J. Dairy Sci. 92:4797-4804
Drake, S.L., K. Lopetcharat, M.A. Drake. 2011. Salty taste in dairy foods:
Can we reduce the salt? J. Dairy Sci. 94:636-645.
Donadini, D., M.D. Fumi, L. Valoni and S. Porreta. 2012. Hedonic
Response to cheese in preschoolers. J. Sens. Studies 27: 176-187.
45
Ennis, D.M. and J.M. Ennis. 2013. Mapping Hedonic Data: a process
perspective. J. Sen. Studies 28:324-334.
Giménez, A., G. Ares and A. Gámbaro. 2008. Consumer perception of
sandness in Dulce de leche. J. Sen. Studies 23:171–185
Granato, D., Calado, V. M . A., Jarvis, B. 2014. Observations on the use of
statistical methods in food science and technology. F. Res. Int. 55: 137-149.
Heymann, H., B. Machado, L. Torri and A.L. Robinson. 2012. How many
judges should one used for sensory descriptive analysis? J. Sen. Studies 27:111–
122.
Kaaki, D., O. Kebbe Baghdadi, N.E. Najm and A. Olabi. 2012. Preference
mapping of commercial Labneh (strained yogurt) products in the Lebanese
Market. J. Dairy Sci. 95:521-532.
Macfie, H. J., N. Bratchell, K. Greenhoff and L. Vallis. 1989. Designs to
Balance the Effect of Order of Presentation and First-order Carry-over Effects in
Hall Tests. J. Sen. Studies 4:129-148
Market Research World. 2013. Understanding Penalty Analysis. Jun. 08,
2014. http://www.marketresearchworld.net/content/view/3233/74/
Meullenet, J.F., R. Xiong and C.J. Findlay. 2007. Multivariate and
Probabilistic Analyses of Sensory Science Problems. 5 th ed. IFT, Blackwell. Iowa,
US
Moskowitz, H. R. 1983. Product Testing and Sensory Evaluation of Foods.
Marketing and R & D Approaches. Food and Nutrition Press Inc., Connecticut, US.
Morais, E. C., A. G. Cruz, J. A. F. Faria, and H. M. A. Bolini. 2014a.
Prebiotic gluten-free bread: Sensory profiling and drivers of liking. Lebenson.
Wiss. Technol. 55:248–254.
46
Morais, E.C., A.R. Morais, A.G. Cruz and H.M.A. Bolini. 2014b.
Development of chocolate dairy dessert with addition of prebiotics and
replacement of sucrose with different high-intensity sweeteners. J. Dairy Sci.
97:2600-2609.
Murtaza, M. A., S. U. Rehman, F. M. Anjum, and N. Huma. 2013.
Descriptive sensory profile of cow and buffalo milk Cheddar cheese prepared
using indigenous cultures. J. Dairy Sci. 96:1380–1386.
Narayanan, P., B. Chinnasamy, L. Jin and S. Clark. 2014. Use of just-
about-right scales and penalty analysis to determine appropriate concentrations of
stevia sweeteners for vanilla yogurt. J. Dairy Sci. 97:3262-3272.
Perrone, I. T., R. Stephani and B. S. Neves. 2011. Doce de Leite:
Aspectos Tecnológicos. Instituto de Laticínios Cândido Tostes: Juiz de Fora, BR.
Pimentel, T. C., A. G. Cruz, and S. H. Prudencio. 2013. Influence of long-
chain inulin and Lactobacillus paracasei subspecies paracasei on the sensory
profile and acceptance of a traditional yogurt. J. Dairy Sci. 96:6233–6241.
Prinyawiwatkul, W., P. Chompreeda. 2007. Applications of discriminant
and logistic regression analysis for consumer acceptance and consumer-oriented
product optimization study. Pages 279-295 in Accelerating new food product
design and development. J. h. Beckley (Ed), ed. Blackwell Publishing
Professional, Ames, Iowa
Resano, H., A. I. Sanjuán, I. Cilla, P. Roncalés, and L.M Albisu. 2010.
Sensory attributes that drive consumer acceptability of dry-cured ham and
convergence with trained sensory data. Meat Sci., 84:344-351
47
Sensory Society. 2013. Penalty Analysis. Accessed Jun. 08, 2014.
http://www.sensorysociety.org/knowledge/sspwiki/Pages/Penalty%20Analysis.asp
x
Stone, H., R. N. Bleibaum and H. A. Thomas. 2012. Sensory Evaluation
Practices. Elsevier: New York
Vidigal, M.C.T.R., V.P.R. Minim, N. B. Carvalho, M.P. Milagres and A. C.A.
Gonçalves. 2011. Effect of a health claim on consumer acceptance of exotic
Brazilian fruit juices: Açaí (Euterpe oleracea Mart.), Camu-camu (Myrciaria dubia),
Cajá (Spondias lutea L.) and Umbu (Spondias tuberosa Arruda). Food Res. Int.
44:1988-1996
Xu, L., Cai. C.B., Cui, H.F., Ye, Z.H., and Yu, X. P. 2012. Rapid
discrimination of pork in Halal and non-Halal Chinese ham sausages by Fourier
transform infrared (FTIR) spectroscopy and chemometrics. Meat Sci. 92:506-510
Zalazar, C. A.; Perotti, M. C. 2011. Concentrated Dairy Products, Dulce de
Leche. Pages 874-880 in Enciclopedia of Dairy Science. 3th ed. J. W. Fuquay, P.
Fox, & P. McSweeney (Eds.), ed. Academic Press, Cornwall, UK
Zielinski, A. A. F., C. W. I. Haminiuk, C. A. Nunes, E. Schnitzler, S. van
Ruth, D. Granato. 2014. Chemical composition, sensory properties, provenance,
and bioactivity of fruit juices as assessed by chemometrics: a critical review and
guideline. Comprehensive Reviews in Food Science and Food Safety. 13:300-316
48
TABLES AND FIGURES
49
50
51
52
53
54
55
56
57
Figure 1. Principal component Analysis (PCA) of Dulce de Leche samples,
F1 x F2, of attributes and samples (I, II, III, IV, V, VI ,VII), as given in Table 3.
58
Figure 2. External preference mapping illustrating: three clusters, 1 (circular,
“(+)“), 2 and 3 (saddle, “(o)”); samples (I, II, III, IV, V, VI and VII); and the five regions of the global average value of acceptance.
59
3.2 ARTIGO 2: DULCE DE LECHE, A TYPICAL PRODUCT OF LATIN AMERICA: CHARACTERIZATION BY PHYSICOCHEMICAL, OPTICAL AND INSTRUMENTAL METHODS
60
Dulce de Leche, a typical product of Latin America: characterization by
physicochemical, optical and instrumental methods
L.V. Gazea, M.P. Costaa, M.L.G. Monteiroa, J.A.A. Lavoratoc, ,
C. A.Conte Júniora, A.G Cruzb , M.Q. Freitasa*
a Universidade Federal Fluminense (UFF), Faculdade de Veterinária,
Pós-Graduação em Higiene Veterinária e Processamento de Produtos de
Origem Animal, 24230-340 Niterói, Rio de Janeiro, Brazil
b Instituto Federal de Educação, Ciência e Tecnologia do Rio de Janeiro
(IFRJ), Mestrado Profissional em Ciência e Tecnologia de Alimentos (PGCTA),
Rua Senador Furtado, 171/175, Maracanã, 20270-021 Rio de Janeiro, Brazil
c LabCaseus (Analysis Lab Water and Food), Leopoldina, Minas Gerais,
Brasil CEP 36.700-000.
*Corresponding author:
M.Q. Freitas
Rua Vital Brazil Filho, nº 64
Santa Rosa - Niterói, Rio de Janeiro, Brasil
CEP: 24.230-340
Phone.: +55 - 21 - 2629-9545
E-mail: [email protected] (M.Q. Freitas)
61
Abstract
The physicochemical profile of Dulce de Leche (DL) was determined by both
routine analysis and others techniques (HPLC, GC-MS and ICP-OES). Seven
Brazilian commercial brands were characterized for moisture content, protein, fat,
ash, pH and titratable acidity, mineral content (sodium, potassium, calcium, and
phosphorus), optical parameters and instrumental analysis (carboidrates content and
volatile compounds). Overall, extensive variability among all the parameters
evaluated were observed, suggesting different operational procedures in the dairy
factories along the DL processing. In this sense, an increase of intrinsic quality of DL
is related closely the standardization of operacional parameters using during the
manufacture.
Key-words: dulce de leche, quality, physico-chemical analysis.
62
1. Introduction
Dulce de Leche is a concentrated milk product obtained by heat treatment
with or without negative pressure, with ingredients added, especially sucrose
conferring differential sensory and physicochemical characteristics as compared to
other dairy products (Velasco, Quezada, Parra, Campos, Villalobos & Wells, 2010;
Perrone, Stephani & Neves, 2011; Ranalli, Andrés & Califano, 2012). It is a highly
consumed product (Giménez, Arés & Gambáro, 2008; Oliveira, Penna & Nevarez,
2009) in Latin America where it is traditionally made, mainly in Argentina and
Uruguay, followed by Brazil, Chile, Paraguay and Bolivia (Zalazar; Perotti, 2011).
Additionally, some countries as Brazil have exported DL to other economic blocs
such as European Union (EU) and Asia Pacific Economic Cooperation (APEC),
evidencing the expansion process of the export market of Dulce de Leche (MDIC,
2014).
Although DL is composed of milk, sucrose, sodium bicarbonate and
other additives, it may exhibit distinct characteristics that vary greatly among
industries and production areas. Some factors such as management of cows,
genetics, good agricultural practices, and formulation and final processing can affect
DL characteristics (Smit, 2000). Different processing parameters including heating
time and temperature, intensity of negative pressure, and mass balance have direct
influence on the physicochemical and sensory attributes of the final product (Smit,
2001; Perrone, 2011). These characteristics are strongly related to nutritional,
technological and sensory quality of DL, consequently affecting the consumers
purchase intention, opening new markets, reducing production costs, and improving
quality control (Giménez, 2008). Thus, the physicochemical characteristics of the DL
should be compulsorily monitored.
63
However, currently there are few studies on the chemical composition
of the DL, and there is a lack of information on its chemical profile using instrumental
techniques such as Gas Chromatography Mass - Spectrometry (GC-MS), Inductively
Coupled Plasma Optical Emission Spectrometry (ICP-OES) and High Performance
Liquid Chromatography (HPLC). (Oliveira, 2009). In this context, further studies on
the physicochemical characteristics using these techniques are needed.
Under this view, the aim of this study was to characterize different
commercial DL brands by conventional methods and other instrumental techniques
such as HPLC, GC-MS and ICP-OES, contributing to relevant information about a
product with increasing demand and great relevance in terms of Latin American
economy.
2. Materiais and methods
2.1 Sampling
Seven DL samples were purchased in Brazilian markets. All brands
were produced in processing plants under inspection by the Federal Inspection
Service (SIF), being marketed throughout the Brazilian territory. The transport to the
laboratory was done at room temperature in original containers. For the analyses, the
mean values were calculated by using three analytical replicates, except for the
optical characterization, performed with 10 replicates.
2.2 Proximate composition, acidity and pH
The proximate composition (moisture, lipid, protein and ash) was
determined according to the recommendations of the Association of Official
Analytical Chemists (AOAC, 2005). The pH and the percentage of lactic acid were
determined according to Adolfo Lutz Institute methodology (IAL, 2008). Moisture was
determined gravimetrically by oven drying (Micronal, São Paulo, Brazil) until constant
weight. The ash content was determined gravimetrically after incineration of 3 g
64
sample at 550 °C in a muffle furnace. Protein content was calculated using the total
nitrogen determined by Kjeldahl method with subsequent multiplication by the factor
6.38. The fat was determined using soxhlet extraction method after denaturation of
proteins and carbohydrates with hydrochloric acid. The pH was measured in digital
potentiometer (MICRONAL B-375) by direct insertion of the electrode in 10g of the
samples. The lactic acid was mensured by titration with sodium hydroxide using
phenolphthalein as indicator.
2.3 Sodium, Calcium, Potassium and Phosphorus by ICP-OES
Sodium, calcium, potassium, and phosphorus were determined by
mineralization in a microwave oven, followed by quantification in ICP-OES
Spectrometry (Spectro Analytical Instrument – Spectroflame P) (AOAC, 2005).
Sample digestion (1-2 g) was carried out by acid hydrolysis, using 2 ml nitric
perchloric acid solution (2:1) for approximately 16h at room temperature. After
oxidation, the samples were heated in a digestion block (Tecnal, São Paulo, Brazil) in
a fume hood at low boil to 100 °C (± 2 °C) for 1h, and maintained for additional 2h at
170 °C (±2 °C). The mineral content of the Dulce de Leche samples was determined
by Inductively Coupled Plasma Optical Emission Spectrometry (ICP-OES). The
micronutrient concentrations were calculated based on calibration curves with
sodium, calcium, potassium and phosphorus standard solutions.
2.4 Carbohydrates by HPLC
The analysis of carbohydrates (lactose, sucrose, and glucose) was
performed by high performance liquid chromatography (HPLC) according to the
extraction methodology described by Llano, Rodriguez & Cuesta (1996) with some
modifications. Firstly, 1g DL was weighted in a 25 mL Becker and dissolved with 5
mL sulfuric acid (45 mmol.L-1). Then, the sample was transferred to a conical tube,
65
homogenized in a shaker (Certomat® MV, B. Braun Biotech International, Melsungen,
Germany) for 1 min at 2500 rpm, and stirred for an hour on a shaker table set at 250
rpm. Subsequently, it was centrifuged at 1.000g for 30 min at 4°C and the
supernatant was filtered on filter paper Whatman # 1. The filtered samples were
injected (20μL) into the liquid chromatograph (Shimadzu®, Japan) Model LC/20 AT,
equipped with integrator CBM-20A, and with detector RID-10A kept at 40ºC using a
cation-exchange Aminex HPX-87H column 300 x 7.8 mm (Bio-Rad, Hercules, CA,
USA) maintained at 60°C.
The mobile phase was water (Millipore Corp., Billerica, MA) at a flow
rate of 0.5 mL min-1 (Isocratic). HPLC chromatograms were obtained using LC
Solution software (Shimadzu Corp.). Quantitation was performed by external
standard method using lactose, glucose and sucrose (Sigma-Aldrich, St. Louis,
Missouri, USA) as standards at the following dilutions: 6.25, 10, 12.5, 15, 20, 25, 30,
35, 50, and 75 mg.mL-1. Under these conditions, sucrose showed a double peak that
coeluted with the glucose in the first peak. Thus, the glucose peak was measured by
difference using the ratio between the major and minor sucrose peaks
2.5 Volatile compounds by GC-MS
The extraction and identification of volatiles were made by solid phase
microextraction (SPME) and by gas chromatography coupled to mass spectrometry
(GC-MS), respectively, according to methodology described by Condurso, Verzera,
Romeo Ziino & Conte (2008). The volatile compounds were extracted by the static
headspace method. SPME was performed with a commercially available fibre
housed in its manual holder (Supelco, Bellefonte, PA, USA). All extractions were
carried out using a DVB/CAR/PDMS (divinylbenzene / carboxen /
polydimethylsiloxane) fiber, 50/30 μm film thickness (Supelco, Bellefonte, PA, USA).
66
For that, 10 g sample was dissolved in 12 mL saturated NaCl solution, keeping the
vial at 40ºC, with equilibrium time of 20 min and extraction time of 30 min. After
sampling, the SPME fibre was introduced onto the GC-MS injector, kept in the
splitless injector, and maintained at 260 ºC for 3 min for thermal desorption of the
analytes. To identify the components, the linear retention index (LRI) of the
experimental spectra was calculated and compared to LRI of NIST'98
(NIST/EPA/NIH Mass Spectra Library, version 1.7, USA) (Van den Dool and Kratz,
1963).
2.6 Optical Characterization
The color parameters were determined using Chroma meter (CR-410,
Konica Minolta Sensing, Inc., Tokyo, Japan) using iluminant D65, previously
calibrated. All measurements were performed at 8 °C. The results were expressed in
L* (lightness; 0 = black, 100 = white), a* (+ a* = redness, - a* = greenness) and b* (+
b* = yellowness, - b* = blueness) values. Chroma (C*) was calculated using the color
coordinates a* and b*. These attributes indicate the color intensity of the samples by
the degree of perceived hue as compared with the gray tone with the same lightness.
To identify a certain colour dfference with reference to grey colour with the same
lightness, the hue angle (h*) was calculated. The total color difference (ΔE*) was
used to compare one sample to the other. To better guide the discussion, total color
difference, (Δa*2 + Δb*2 + ΔL*2)1/2 was classified as very distinct (>3), distinct (1.5<
ΔE<3) and small difference (<1.5) (Pathare, Opara, Al-Said, 2013).
2.7 Statistical Analysis
The results of the physicochemical analyses were subjected to one-way
ANOVA, considering the sample as fixed effects. The differences between means
67
were submitted to Tukey’s test at 95% confidence level. In addition, Pearson’s
correlation was performed to appreciate and interpret interactions between the
variables through the linear correlation coefficient (Granato, Calado & Jarvis, 2014).
Principal Component Analysis (PCA) was performed using with the means
values of the physicochemical and optical parameters (Aquino, Silva, Freitas, Felicio,
Cruz, & Conte-Junior, 2014). The matrix data set was composed of 7 lines and 19
columns, being the former the samples and the latter the values of the different
analyses. The data were auto scaled before the analysis. The statistical package was
XLSTAT software version 2013.2.03 (Addinsoft, Paris, France).
3.Results and Discussion
3.1 Chemical composition, acidity and pH
Results of proximate composition, pH and titratable acidity of the DL
samples are in Table 1 . Protein, lipid, ash, moisture, lactic acid content and pH
ranged respectively from 3.51 to 7.12; 3.56 to 6.99; 1.31 to 2.05; 17.49 to 29.67; 0.23
to 0.50; and 6.14 to 6.37 w/w (p < 0.05), with a significant variability in all
parameters. These results are in agreement with other studies on DL (Demiate,
Konkel & Pedroso, 2001; Ranalli, 2012). A strong correlation between lipid and ash
contents (r = 0.89, p <0.008) was observed. Once the lipids of DL comes only from
milk (BSR 1997), probably the ash comes from milk despite a small portion may be
added to the formulation by the addition of sodium bicarbonate (Perrone, 2011).
Therefore, a high percentage of these compunds may indicate high
representativeness of milk in the Dulce de Leche formulations.
Faced with the need to establish standards for DL, the Mercosur
countries have created an identity and quality regulation (BSR, 1997), establishing
the following limits for the pasty DL: moisture (max 30 g / 100 g), fat (6 – 9 g / 100 g),
68
ash (max 2 g / 100 g) and protein (up 5 g /100 g). Therefore, low protein and lipids
contents were found in the present study (samples II, IV, V and VII). The deficit of
proteins and lipids compromises the nutritional value of the product, suggesting the
replacement of milk by water or carbohydrates such as sucrose and starch. Indeed,
food fraud often has been considered to be foremost an economic issue and less a
concern of the traditional food safety or food protection intervention and response
infrastructure (Moore, Spink, & Lipp, 2012). However, any adulteration may result in
changes of identity and/or purity of the original and purported ingredient by replacing,
diluting, or modifying it by physical or chemical means.
Regarding the acidity, only samples II and V were statistically (P<0.05)
equal, with values of 0.37 and 0.36, respectively. This is important for sensory
characteristics, since the acid taste has a negative influence on the product
acceptance for masking the sweet taste (Giménez, 2008).
3.2 Sodium, Calcium, Potassium and Phosphorus content
Minerals are important compounds in dairy products (Cashman, 2006).
Sodium, for example, is present in milk (Oliveira, 2009) and has been gaining
importance due to the need for reducing sodium in human diet to prevent
hypertension (Felicio et al., 2013; Tanase, Griffin, Koski, Cooper & Cockell, 2011;
Cruz et al., 2011). In addition, the effect of other minerals including calcium,
potassium and phosphorus in consumer health has been widely investigated
(Tanase, 2011; Childers, Corman, Edwards & Elser, 2011). Since an information gap
remains in the recent literature, instrumental analytical methods can be used to
identify and quantify these minerals, including the mineralization technique by
69
microwave cavity and Inductively Coupled Plasma Optical Emission Spectrometry
(ICP-OES).
Table 1 shows the concentrations of sodium, calcium, potassium and
phosphorus in Dulce de Leche samples. A significant difference between these
parameters was observed for all samples, ranging from 0.12 to 0.16 g / 100 g; 0.28 to
0.40 g / 100 g; 0.19 to 0.36 g / 100 g; 0.14 to 0.24 g / 100 g for sodium, calcium,
potassium and phosphorus, respectively (p <0.05), with the highest differences
observed for sodium and calcium (p <0.05).
Sodium bicarbonate acts as a processing adjuvant or as a sodium
based preservative, such as citrate allowed by BSR (1997). Another possibility for the
high sodium levels could be salt imbalance in cattle feed, or a loss of homeostasis of
their mammary glands, probably due to health issues, producing milk with low
mineral content (Ferreira, Bislevc, Bendixenc & Almeida, 2013). To discuss better,
the correlation of sodium and ash content must be investigated, since ash represents
a set of all minerals found in milk.
A moderate correlation was observed between ash content and sodium (r =
0.53, p <0.219) as compared to the correlations with potassium (r = 0.98, p <0:00),
calcium (r = 0.99, p <0.0001), and phosphorus (r = 0.97, p <0.000), with coefficients
close to 1.00, indicating strong positive dependence between these variables. The
same strong correlation is obtained when potassium, phosphorus and calcium are
compared with each other - sodium x phosphorus (r = 0.97, p <0.000), potassium x
calcium (r = 0.99, p <0.0001), and calcium x phosphorus (r = 0.98, p <0.000).
Therefore, it can be deduced that the presence of sodium in the different DL samples
may be related to its addition during the manufacturing process as a technological
adjuvant, rather than the physiology of dairy cattle influencing the raw material. It
70
should be taken into account that the World Health Organization (WHO, 2006)
recommends the consumption of less than 2.0 g sodium chloride per day to prevent
the incidence of hypertension. Once the DL contains low sodium levels (about 0.14
g/100 g), its consumption does not significantly contribute to the occurrence of
hypertension even in moderate amounts, since consumer should eat unreal amounts
of DL (at least 1.42 kg / day) to meet WHO requirements. In the sensory point of
view, the changes in sodium levels do not alter flavor and / or aftertaste (Hough,
Bratchell & Macdougall, 1992b), but can influence color, which in turn may affect
consumers acceptance (Hough, Bratchell & Wakeling, 1992a).
From a nutritional standpoint, calcium is the most important element of
milk, being fundamental for healthy bones and osteoporosis prevention. Calcium
levels in milk do not vary greatly (Smit, 2000; Cashman, 2006). Moreover, since the
dry extract is concentrated during the manufacturing process, calcium, potassium,
and phosphorus are interesting attributes to indicate the representativeness of the
raw material in DL formulations.
3.3. Carbohydrates by HPLC
Among the DL components, carbohydrates have greater importance due to
its influence on technological processing. The two most important carbohydrates in
DL are lactose and sucrose (Demiate, 2001; Oliveira, 2009). Lactose is present in
milk products without added β-galactosidase or acid-lactic cultures (Giménez, 2008;
Oliveira, 2009; Vénica, 2013), and is responsible for about 50% of the osmotic
pressure due to its concentration (4.8%) in the fluid phase (Smit, 2000). A large
amount of sucrose is added to DL formulations during manufacturing (Perrone,
2011), which is responsible, on a larger scale than lactose by the non-enzymatic
browning during heat treatment, and by providing the characteristic sweet taste of DL
71
(Velasco, 2010; Perrone, 2011; Golon & Kuhnert, 2012; Jiang, Wang, Wu & Wang,
2013). Moreover, it is a common practice adding glucose to aid the browning
process, to promote greater brightness, and let the product pleasing to the palate
(Oliveira 2009). Therefore, it can be concluded that the DL is a typically brownish
product (Pathare, 2013).
Rapid methods for quantification of carbohydrates in processed foods have
been developed (Parc, Lee, Chen & Barile, 2014). Quantification and
characterization methods for DL have been reported and adapted by enzymatic and
photometric methods (Demiate, 2001; Guimarães, Leão, Pimenta, Ferreira &
Ferreira, 2012). However, recent literature studies have focused on quantifying
carbohydrates in dairy foods by high performance liquid chromatography (HPLC),
which is commonly used for other dairy products such as milk (Giménez, 2008; Erich,
Anzmann & Fischer,, cheese (Gomes et al ., 2011) and yogurt (Cruz et al., 2012a,b;
Cruz et al., 2013),
Table 2 shows lactose, sucrose and glucose contents of Dulce de
Leche samples. Again, differences were observed for all sugars and for total
carbohydrates between samples (p <0.05). The samples I and III showed the highest
lactose content (11.75 and 13:51/ 100 g, respectively) while samples II, III and IV
presented the highest sucrose values (49.59, 47.40, and 47.36 g/ 100 g,
respectively). With respect to glucose, the samples II, V and VII presented the three
highest values (0.40, 0.25 and 0.75 g/ 100 g, respectively). The total carbohydrates
ranged from 61.08 to 43.67 g/100 for samples III and V, respectively. A great
variation was observed for glucose content, which ranged from 0.09 to 0.75 g /100 g
for the sample I and VII, respectively, suggesting heterogeneity of the technological
processing of DLs.
72
Lactose has a strong correlation with most milk components, such as protein
(r = 0.82, p <0.023), fat (r = 0.93, p <0.003), ash (r = 0.89, p <0.008), potassium (r =
0.93, p <0.003), calcium (r = 0.90, p <0.006) and phosphorus (r = 0.87, p <0.010). A
strong correlation between the three carbohydrates (total) and sucrose was observed
(r = 0.95, p <0.001), thus demonstrating the importance of sucrose in glycide profile
of DL and hence the characterization of this product. The large amount of sucrose
and lactose found in sample III increased its susceptibility to non-enzymatic browning
reactions and especially lactose crystallization. Sample III was the only sample that
exhibited perceptible crystals while still within the shelf life. This can be explained by
the fact that the addition of sucrose is inversely proportional to the solubility of
lactose. To avoid this, the industry can add glucose syrup for its influence on the
plasticity, which allows obtaining higher solids content without reaching the saturation
point of lactose (Oliveira, 2009). Another strategy to avoid crystallization by
decreasing solubility in traditional DL is to reduce the non-hydrolyzed lactose content
in relation to other saccharides, as observed in the sample II.
Regarding the sweet taste, only samples V and VII did not present this
parameter with high intensity, due to the high sweetening effect of sucrose when
compared to other carbohydrates. To harmonize this situation, glucose as a reducing
sugar, has the ability to optimize browning at high temperatures and generate
caramel and bitter compounds, masking the too sweet taste. Besides the high
temperatures, increasing cooking time is directly proportional to the generation of
complex compounds that influence flavor and residual sensation (Hough, 1992b).
Nevertheless, Hough (1992a) reported that the flavor has a low impact on the overall
impression.
73
The correlation between sucrose and moisture content (r = -0.59, p
<0.158) highlights the importance of balance between both concentrations. The high
levels of sucrose have beneficial technological effects, such as reducing water
activity, reducing microbial growth and increasing shelf life (Oliveira, 2009). Sample II
was manufactured in a food exporting industry, therefore it is interesting to extend
their validity due to the long distances to reach overseas markets and the shelf life
period. Among all samples, the sample II exhibited greater shelf life shown on the
label, corroborating the low moisture and high sucrose content found in this sample.
The analytical techniques used in this study were not intended to
quantify the starch.Nevertheless, it is possible that starch had been added to the DL
formulations containing low sucrose, glucose and lactose contents (samples II, V and
VII) in order to supplement their glycidic matrix, since the addition of this component
was informed on labels of these brands. Demiate (2001) also suggested that whey
protein could be added. Despite the addition of sucrose and starch is permitted, it
decreases protein, lipid and ash contents, affecting negatively the nutritional
composition (Demiate, 2001; Konkel, Oliveira, Simões, Demiate, 2004). Therefore,
starch addition should be evaluated carefully by the surveillance authorities.
Glucose added at the end of the manufacturing process is capable of forming
a strongly hydrated complex with protein, which increases the DL viscosity and
directly interfering with the formation of lactose crystals perceptible in the mouth
(Giménez, 2008; Perrone, 2011). The replacement of sucrose by glucose also results
in higher brightness, thin, soft and smooth texture, thus increasing consumer’s
acceptance. Opposite attributes are observed when high levels of glucose syrup are
added to the formulations (Hough, 1992a; Hough, 1992b). Therefore, the addition of
this important sugar should be done at the end of the process, when the temperature
74
tends to decrease, in order to prevent drastic changes in DL characteristics (Perrone,
2011). As a reducing sugar, glucose is more reactive than sucrose. Then, at pH 5-6,
there is an active participation of this monosaccharide in the Maillard reaction,
increasing the tendency to browning (Oliveira, 2009). This is an important factor,
since the acceptance of DL is related to its appearance and texture (Hough, 1992a).
Additionally, texture attributes such as hardness, viscosity, and tooth packing are
related to the formation of caramel (Steiner, Foegeding & Drake, 2002). Although no
abrupt changes in color attributes were observed, higher brightness was observed in
the three samples containing higher contents of glucose (samples II, V and VII). Their
addition was probably for this purpose.
The presence of glucose in the other samples can be explained by two
different ways: first, it may have been added without informing the consumer and the
regulatory agencies that approve the labels; second, and more likely, it is related to
the formation of this monosaccharide from the sucrose inversion, releasing glucose
and fructose. This could easily occur in the DL due to the technological processing at
high temperatures. To a lesser extent, it could be related to the presence of minerals
or deficient alkalization during manufacturing process (Oliveira, 2009). Similar
situation was observed by Demiate (2001), who identified glucose in 37 Brazilian
trends, of which only 26 have this carbohydrate reported on the label.
Corroborating BSR (1997), Perrone (2011) suggested a mass balance of a
hypothetical DL manufactured from milk with proximate characteristics found in
Brazil. The DL formulation should have 70 g/ 100 g total solids (30 g/ 100 g
moisture), and processed from 1000L of milk containing 12.21 g/ 100 g total solids
(87.79% water, 4.8% lactose, 3.3% lipids, 3.3% protein, and 0.81% ash), and with
addition of 200 kg sucrose. Thus, 460.14 kg of DL would be produced, presenting all
75
the physicochemical parameters (30 g /100 g water, 7.17 g /100 g fat, 7.17 g / 100 g
protein, and 1.76 g / 100 g ash) established by Brazilian regulation. Furthermore, the
final lactose (10.43 g / 100 g) and sucrose content (43.47 g /100 g) was calculated.
Comparing these two glycidic levels, as seen in Table 2, there is an overestimation of
the use of sucrose in the samples of this study.
The feasibility of measuring lactose, sucrose and glucose separately
suggests that the current legislation (BSR, 1997) should be reviewed. Currently, the
sucrose addition is still measured in fluid milk, which complicates the measurement of
the amount added, since it would be necessary to know the yield to calculate the
mass of sucrose added per liter of milk. Thus, it is suggested that the quantification
and establishment of sucrose limits may be directly in the final product.
Similarly, other mono- and disaccharides should be investigated and studies
on its characterization in DL should be performed. Due to the strong correlation
between lactose and milk components (protein, fat, and ash), based on the results of
the samples III and VI, which fully met the current legislation parameters (BSR,
1997), and on the calculation performed by Perrone (2011) (DL containing 10.43 g/
100 g lactose), it can be stated that the most true lactose values were found in these
samples (13.51 and 11.76 g / 100 g respectively), and on the calculation of Perrone.
3.4 Volatile compounds by GC-MS
The gas chromatography - mass spectrometry (GC-MS) is valuable in
identifying volatile components in food systems, since it provides the linear retention
index (LRI) of the compounds retained on the column (Condurso, 2008). These
volatile compounds of the food matrix may be derived from chemical reactions
involving various factors such as bacterial enzymes, heat treatment, or oxidation
reactions in the presence of atmospheric oxygen. Therefore, their identification is
76
important to verify desirable or undesirable changes that may occur in the product
during processing or storage, which can result in loss of physicochemical and
sensory characteristics, affecting the shelf life of the product. (Smit, 2000).
The volatile compounds and their respective retention indices identified in the
Dulce de Leche samples are shown in Table 3. Overall, 32 volatile compounds were
identified belonging to 10 different chemical families of which only two were not
present in all samples. The compound 4-hydroxy-4-methyl-2-pentanone (diacetone
alcohol) was identified only in the sample VII, while the sorbic acid was detected in
sample II. The former is known to be product of the degradation of milk poly-
unsaturated fatty acids (PUFA) and it can be related with specific temperature used
by this producer along the heat treatment (Bendall, 2011), while the latter is a known
preservative used to prolong the shelf life of processed foods, with effective action
against molds and yeasts. It is relevant to mention that potassium sorbate, which
precedes sorbic acid, was not declared only in I and II labels. So, DL I is
nonconforming under this point of view.
The flavor and aroma of cooked milk is related to dimethyldisulfides. These
sulfur-containing compounds tend to be powerful odorants. Furans and lactones
were also identified in the samples of the present study. These substances, which
are associated with the Maillard reaction, confer flavor and aroma related to the
sterilization process at high temperatures (Zabbia, Buys, & Koch, 2012) In general,
the presence and the level of ketones is directly proportional to the "age "of the
product, compromising its sensory quality, since it may lead to rancid flavor and
aroma (Smit, 2000). Furfural is a known compound, generated from the non-
enzymatic browning, mainly by the caramelization reactions (Paravisini et al., 2012).
3.5. Optical analyses
77
As shown in Table 4, significant differences (p< 0.05) were observed for all
samples with respect to the optical parameters. The parameters L* and h* varied
similarly between samples, which can be supported by the strong correlation
between them (r = 0.93, p <0.003). In the present study, the sample I showed the
highest lihtness (62.11) and Hue angle (62.17), while the sample VI had the lowest
values for these parameters (48.59 and 51.97 respectively), tending to be darker.
The positive values found for the color coordinates a* and b* indicate that the DL is a
product with a tendency to red and yellow, with a predominance of the second, due
to higher b* values found in all samples. Therefore, the combination of these two
colors can explain the color of the DL after the induced browning by the Maillard and
caramelization reactions (Golon, 2012; Pathare, 2013). The parameter a* ranged
from 13.60 to 16.90 in the samples III and V, respectively. The samples VI and V had
the lowest (17.89) and the highest (27.70) b* values, respectively. The differences
among the samples with respect to the color parameter are probably due to the
differences in protein and sugar composition, as well as the changes in time,
temperature and pressure, according to the protocol of each industry (Oliveira, 2009).
This argument can be supported by the h* values, in which the variation of 51.97
(sample VI) to 62.17 (sample I) represents the change, in degrees, from red to
yellow. Therefore, this values (larger then 45º) illustrate the tendency of DL to be
yellowish. Chroma was more intense (32.45) in the sample V and less intense
(22.71) in the sample VI. Although a strong correlation between C* and b* (r = 0.99, p
<0.0001) was observed, it may be because b* values were higher than a* values,
thus increasing C* with more efficiency.
The samples VI and IV, and I and II had the lowest and the highest L* values,
respectively. With respect to b*, only the sample I did not present the highest value.
78
The other three samples presented the same behavior as observed for L*. These
dissimilarities are reflected in the total color difference observed between these
samples: I x IV (11.90), I x VI (16.33), II x IV (10.20) and II x VI (14.73). In addition, V
x VI had high ΔE* (13.07), once V presented the highest a* (16.90) and b* (27.70)
values, while the sample V exhibited the lowest L* (48.59) and b* (17.89). Only the
samples I x II (2.32) and III x VII (1.70) were not considered "very different" to each
other to the human eyes, according to Pathare (2013).
3.6. Principal component analysis
Figures 1 and 2 (a and b) shows the principal component analysis
performed with the data of physicochemical and optical analyses of the DL samples.
Three main components (F1, F2, F3) were used, which accounted for 84.05% of data
with a range of 51.45 and 17.89% for the first and second component, respectively,
while the third main component contributed to 14.73 % of the data. The results
suggest that there is a wide variability in the operating parameters of the DL
processing, which are intrinsic of formulations in the processing units. In a broader
analysis, this study indicates the need to establish a quality standard for the Dulce de
Leche in order to standardize the products available in the market.
F1 is associated positively with the DL physicochemical parameters
comprising the dry extract, except for glucose, which stood on the negative quadrant.
F2 highlighted the pH and the visual attributes, except a*. Through these two
components, it is possible to separate the samples into three groups. Samples III, IV
and VI are allocated on the positive quadrant of F1, identified by most of the DL
physicochemical characteristics related to the composition of the raw material, with
the exception of sucrose and sodium. The negative quadrant of F1 generated a
second group consisting of samples V and VII, highly influenced by the high glucose
content of the samples. Based on the optical characteristics, F2 cooperated with the
79
formation of a third group consisting of samples I and II. The visual aspects are
important to the product acceptance, and can vary in brightness, color and visual
texture according to the process adopted (Giménez, 2008; Velasco, 2010; Perrone,
2011).
The F3 component is negatively associated with moisture content and optical
parameters, including a*. Glucose, positioned on the positive quadrant of F3, was
important in the characterization of sample VII, since this sample contained higher
levels of this monosaccharide in relation to the others. Visually, the sample VII had
higher brightness, confirming the influence of the vector "glucose" on this sample
(Perrone, 2011). The negative quadrant of F3 was responsible for completely
separating the sample V from VII, due to both its differential color characteristic and
high moisture (V), despite not having extrapolated the standard allowed by law (BSR,
1997).
4. Conclusion
This study successfully evaluated the physicochemical characterization of the
pasty DL by conventional methods and instrumental techniques (HPLC, GC-MS, ICP-
OES), providing results that are not yet available for this product.
Thirty-two volatile compounds were identified by GC-MS. With respect to the
sugars, sucrose proved to be the most predominant in this food matrix as compared
to lactose, glucose and other components. Regarding the physicochemical analyses,
four commercial brands showed physicochemical parameters in disagreement to the
Brazilian regulations, due to the low proportion of milk components, primarily lipids,
proteins and ash.
80
There is a need of periodic studies covering the DL intrinsic parameters of
quality which should receive special attention, as it can be directly related to the
consumer acceptance. The malicious or accidental replacement of the dairy
substrate by carbohydrates evidenced by the low-protein, lipid and ash, concurrently
with the moisture within the recommended by law, is a topic that should be
addressed by the industry in the quality control of the final product. Studies should be
performed in this food typical of Latin America, with a view to establish strong identity
and quality standards with the ultimate goal of reaching consumer markets in
different countries.
81
Acknowledgments
The authors are thankful for the financial support of the State of Rio de
Janeiro Carlos Chagas Filho Research Foundation (FAPERJ), process number E-
26/111.525/2013, and the National Council for Scientific and Technological
Development (CNPq), process number 311361/2013-7. L.V. Gaze was supported by
the Coordination for the Improvement of Higher Education Personnel (CAPES).
82
References
Aquino, L.F.M.C., Silva, A.C.O., Freitas, M.Q., Felicio, T.L., Cruz, A.G., &
Conte-Junior, C.A. (2014). Identifying cheese whey an adulterant in milk: Limited
contribution of a sensometric approach. Food Research International, 62, 233-237
Association of Official Analytical Chemists (2005). Official Methods of Analysis
of AOAC International (18th ed.). Gaithersburg, MD: AOAC International.
Bendall, J.G. (2001) Aroma compounds of fresh milk from New Zealand cows
fed different diets. Journal of Agriculture and Food Chemistry 49, 4825-4832.
BSR (1997) Regulamento Técnico MERCOSUL Para Fixação de Identidade e
Qualidade de Doce de Leite. Portaria n◦354, de 04 de setembro de 1997, do Ministério
da Agricultura, Pecuária e Abastecimento, Brasilia.
Cashman, K. D. (2006). Milk minerals (including trace elements) and bone
health. International Dairy Journal, 16, 1389–1398
Childers, D. L., Corman, J., Edwards, M., & Elser, J. J. (2011). Sustainability
challenges of Phosphorus and Food: Solutions from Closing the Human Phosphorus
Cycle. Bio Science, 61, 117–124
Condurso, C., Verzera, A., Romeo, V., Ziino, M.,& Conte, F. (2008). Solid-
phase microextraction and gas chromatography mass spectrometry analysis of dairy
product volatiles for the determination of shelf-life. International Dairy Journal, 18, 819–
825
Cruz, A.G, Faria, J. A. F., Pollonio, M. A. R., Bolinia, H. M. A., Celeghini, R. M.
S., Granato, D., & Shah, N. P. (2011). Cheeses with reduced sodium content: Effects on
functionality, public health benefits and sensory properties. Trends in Food Science &
Technology, 22, 276-291
Demiate,I. M, Konkel, F. E., & Pedroso, R. A. (2001). Avaliação da Qualidade
de Amostras Comerciais de Doce de Leite Pastoso – Composição Química. Ciência e
Tecnologia de Alimentos, 21, 108-114
83
Erich, S., Anzmann, T., & Fischer, L. (2012). Quantification of lactose using ion-
pair RP-HPLC during enzymatic lactose hydrolysis of skim milk. Food Chemistry, 135,
2393–2396
Felicio, T. L., Esmerino, E. A., Cruz, A. G., Nogueira, L. C., Raices, R. S. L.,
Deliza, R., Bolini, H. M. A, &. Pollonio, M. A. R (2013). Cheese. What is its contribution
to the sodium intake of Brazilians? Appetite, 66, 84-88
Ferreira, A. M., Bislevc, S. L., Bendixenc, E., & Almeida, A. M. (2013). The
mammary gland in domestic ruminants: A systems biology perspective. Journal of
proteomics, 94, 110-123
Giménez, A., Arés, G., Gambáro, C. (2008). Consumer reaction to changes in
sensory profile of dulce de leche due to lactose hydrolysis. International Dairy Journal,
18, 951–955
Golon, A., & Kuhnert, N. (2012). Unraveling the Chemical Composition of
Caramel. Journal of Agricultural and Food Chemistry, 60, 3266−3274
Granato, D., Calado, V. M. A., & Jarvis, B. (2014). Observations on the use of
statistical methods in Food Science and Technology. Food Research International, 55,
137–149
Guimarães, Í. C. O., Leão, M. H. M. R., Pimenta, C. J., Ferreira, L. O. &
Ferreira, E. B. (2012). Development and description of light functional Dulce de Leche
with coffee. Ciênc. agrotec., 36, 195 -203
Hough, G., Bratchell, N., & Wakeling, I. (1992a). Consumer preference of Dulce
de Leche among students in the United Kingdom. Journal of Sensory Studies, 7, 119-
132
Hough, G., Bratchell, N., & Macdougall, D. (1992b). Sensory profiling of Dulce
de Leche, a dairy based confectionary product. Journal of Sensory Studies, 7, 157-178.
84
Institute Adolfo Lutz. (2008). Métodos Físico-Químicos para Análise de
Alimentos (4th ed.). São Paulo, Brazil: Instituto Adolfo Lutz.
Jiang, Z., Wang, L., Wu, W. & Wang, Y. (2013). Biological activities and
physicochemical properties of Maillard reaction products in sugar–bovine casein peptide
model systems. Food Chemistry, 141, 3837–3845
Konkel, F. E., Oliveira, S. M. R., Simões, D. R. S., Demiate, I..M. (2004).
Avaliação sensorial de doce de leite pastoso com diferentes concentrações de amido,
Ciênc. Tecnol. Aliment.. 24, 249-254
Llano, D. G, Rodriguez, A. & Cuesta, P. (1996). Effect of lactic starter cultures
on the organic acid composition of milk and cheese during ripening-analysis by HPLC,
Journal of Applied Bacteriology. 80, 570-576
MDIC (Ministério do Desenvolvimento, Indústria e Comércio Exterior). Sistema
de Análise das Informações de Comércio Exterior, AliceWeb (2014). Url
aliceweb.desenvolvimento.gov.br. Accessed 05.04.14
Moore, J. C., Spink, J., & Lipp, M. (2012). Development and Application of a
Database of Food Ingredient Fraud and Economically Motivated Adulteration from 1980
to 2010. Journal of Food Science, 77, 118-126
Oliveira, M. N., Penna, A. L. B., & Nevarez, H. G. (2009); Production of
Evaporated Milk, Sweetened Condensed Milk and ‘Dulce de Leche. In A. Y. Tamime
(Eds.) Dairy Powders and Concentrated Products. (pp. 149-180). Oxford: Blackwell
Publishing.
Paravisini, L., Gourrat-Pernin, K., Gouttefangeas, C., Moretton, C., Nigay, H.,
Dacremont, C., & Guichard, E. (2012). Identification of compounds responsible for the
odorant properties of aromatic caramel. Flavour and Fragrance. Journal, 27,424–432
Parc, A. L., Lee, H., Chen, K., & Barile, D. (2014). Rapid Quantification of
Functional Carbohydrates in Food Products. Food and Nutrition Sciences, 5, 71-78
85
Pathare, P. B., Opara, U. L., & Al-Said, F. A.. (2013). Colour Measurement and
Analysis in Fresh and Processed Foods: A Review. Food Bioprocess Technology, 6, 36–
60
Perrone, I. T., Stephani, R., & Neves, B. S. (2011). Doce de Leite. Aspectos
Tecnológicos. Juiz de Fora, MG: Instituto de Laticínios Cândido Tostes
Ranalli, N., Andrés, S. C., & Califano, A. N. (2012). Physicochemical and
Rheological Characterization of “Dulce de Leche”. Journal of Texture Studies, 43, 115-
123
Smit, G. (2000). Dairy Processing – Improving Quality: Boca Raton, FL: CRC
Press.
Steiner, A. E., Foegeding, E. A., & Drake, M. (2002). Descriptive analysis of
caramel texture. Journal of Sensory Studies, 18, 277-289.
Tanase, C. M., Griffin, P., Koski, K. G., Cooper, M. J., Cockell, K. A. (2011).
Sodium and potassium in composite food samples from the Canadian Total Diet Study,
Journal of Food Composition and Analysis. 24, 237–243
Van den Dool, H., & Kratz, P. D. (1963). A generalization of the retention index
system including linear temperature programmed gas–liquid partition chromatography.
Journal of Chromatography, 11, 463–471.
Velasco, V., Quezada, J., Parra, C., Campos, J., Villalobos, R., & Wells, G
(2010). Effect of different l-carrageenan and sucrose concentrations on some physical
and sensory characteristics of solid-caramel-milk. Revista Chilena de Ciencias
Agropecuarias, 26, 25-35
WHO (World Heath Organization) (2006). Reducing salt intake in populations.
Paris: World Health Organization.
Zabbia, A., Buys, E. M., & De Kock, H. L. (2012). Undesirable Sulphur and
Carbonyl Flavor Compounds in UHT Milk: A Review. Critical Reviews in Food Science
and Nutrition, 52, 21-30
86
Zalazar, C. A.; Perotti, M. C. (2011). Concentrated Dairy Products, Dulce de
Leche. In J. W. Fuquay, P. Fox, & P. McSweeney (Eds.), Enciclopedia of Dairy Science
(2 ed.) (pp. 874-880). Cornwall, UK: Academic Press.
87
TABLES AND FIGURES
88
89
90
91
92
93
Figure 1. Principal component Analysis (PCA) of Dulce de Leche samples,
F1 x F2, of parameters (a) and samples (I, II, III, IV, V, VI ,VII) (b) as given in Tables
1, 2, 3 and 5.
94
Figure 2. Principal component Analysis (PCA) of Dulce de Leche samples,
F1 x F3, of parameters (a) and samples (I, II, III, IV, V, VI ,VII) (b) as given in Tables
1, 2, 3 and 5.
95
4 CONSIDERAÇÕES FINAIS
A aceitação e a escala do ideal tiveram resultados bastantes similares.
Sendo assim, os DLs menos aceitos foram os que tiveram menos citações como
ideais na escala do ideal. A escala do ideal indicou valores mais fora do ideal (acima
ou abaixo) nos doces que apresentaram valores mais extremos na ADQ. Isso
demonstra uma habilidade dessa metodologia em “semi-quantificar” atributos
utilizando consumidores. A LR e a ANN se mostraram, através da alta acurácia,
sensibilidade e especificidade, modelos adequado para predizer os atributos mais
importantes para o consumidor. O método de PLSR demonstrou baixa aplicabilidade
ao DL sugerindo um comportamento mais complexo, não linear, das características
do DL.
Houve êxito na caracterização físico-química do DL pastoso através de
métodos convencionais e técnicas instrumentais (HPLC, GC-MS, ICP-OES), o que
trouxe à comunidade dados que anteriormente não haviam sido reportados. 32
compostos voláteis foram identificados pelo CG-MS. A sacarose demonstrou ser o
açúcar de maior predomínio nessa matriz alimentar em comparação com a lactose,
glicose e os outros componentes. Quatro marcas comerciais se apresentaram com
parâmetros físico-químicos em desacordo com a regulamentação preconizada
havendo baixa proporção dos componentes do leite – basicamente lipídios,
proteínas e cinzas.
Mais estudos devem ser conduzidos nessa matriz alimentícia típica da
América Latina com vistas a se estabelecer padrões mais nítidos da aceitação e das
características físico-químicas com o objetivo final de se atingir mercados
consumidores em diferentes países favorecendo o crescimento regional da
economia.
96
5 REFERÊNCIAS BIBLIOGRÁFICAS
AQUINO, L.F.M.C., SILVA, A.C.O., FREITAS, M.Q., FELICIO, T.L., CRUZ,
A.G., & CONTE-JUNIOR, C.A.. Identifying cheese whey an adulterant in milk: Limited
contribution of a sensometric approach. Food Research International, v. 62, n. 233-
237, 2014
Association of Official Analytical Chemists (AOAC) (2005). Official Methods
of Analysis of AOAC International (18th ed.). Gaithersburg, MD: AOAC International.
BAHRAMPARVAR, M.; SALEHI, F.; RAZAVI, S. M. Predicting Total
Acceptance of Ice Cream Using Artificial Neural Network. Journal of Food Processing
and Preservation. 2013.
BARDOT, L., L. BOCHEREAU, N. MARTIN AND B. PALAGOS.. Sensory-
instrumental Correlation by Combining Data Analysis and Neural Network
Techniques. Food Quality and Preference. v. 5, n. 159-166, 1994
BAYARRI, S.; CARBONELL, I.; BARRIOS, E. X.; COSTELL, E. Impact of
Sensory Differences on Consumer Acceptability of Yoghurt and Yoghurt-like
Products. International Dairy Journal, v. 21, p.111-118, 2011.
BENDALL, J.G. Aroma compounds of fresh milk from New Zealand cows fed
different diets. Journal of Agriculture and Food Chemistry v. 49, p. 4825-4832, 2001.
BRASIL. Portaria nº354, de 4 de setembro de 1997. Regulamento Técnico
de Identidade e Qualidade do Doce de Leite. Diário Oficial da União, Brasília, DF, p.
19685, 08 de setembro de 1997. Seção 1
BRASIL. Leite e derivados. In: INSTITUTO ADOLFO LUTZ. Métodos Físico-
Químicos para Análise de Alimentos. 4ª Ed. Brasília, Ministério da Saúde. Agência
Nacional de Vigilância Sanitária, Ministério da Saúde, 2008. cap.27, p.823-
881.
97
CADOT, Y.; CAILLÉ, S.; SAMSOM, A.; BARBEAU, G.; CHEYNIER, V.
Sensory Dimension of Wine Typicality Related to a Terroir by Quantitative
Descriptive Analysis, Just About Right Analysis and Typicality Assessment. Analytica
Chimica Acta, v. 660, p. 53-62, 2010.
CASHMAN, K. D.. Milk minerals (including trace elements) and bone health.
International Dairy Journal, v. 16, p. 1389–1398, 2006
CETÓ, X., M. GUTIÉRREZ-CAPITÁN, D. CALVO AND M. DEL VALLE. Beer
classification by means of a potentiometric electronic tongue. Food Chemistry. v.
141, n. 3, p. 2533-2540, 2013
CEVOLI, C., L. CERRETANI, A. GORI, M. F. CABONI, T. GALLINA TOSCHI
AND A. FABBRI. Classification of Pecorino cheeses using electronic nose combined
with artificial neural network and comparison with GC-MS analysis of volatile
compounds. Food Chemistry. v. 129, n. 3, p.1315-1319, 2011
CHILDERS, D. L., CORMAN, J., EDWARDS, M., & ELSER, J. J..
Sustainability challenges of Phosphorus and Food: Solutions from Closing the
Human Phosphorus Cycle. Bio Science, v. 61, p. 117–124, 2011
CONDURSO, C., VERZERA, A., ROMEO, V., ZIINO, M.,& CONTE, F.. Solid-
phase microextraction and gas chromatography mass spectrometry analysis of dairy
product volatiles for the determination of shelf-life. International Dairy Journal, v. 18,
p. 819–825, 2008
CRUZ, A. G., CASTRO, W. F., FARIA, J. A. F., CELEGHINI, R. M. S., LIMA-
PALLONE, J. GODOY, H. T.. Glucose oxidase: A potential option to decrease the
oxidative stress in stirred probiotic yogurt. Food Science & Technology, v. 42, n. 512-
515, 2012
CRUZ, A. G.; CADENA, R. S.; FARIA, J. A. F.; OLIVEIRA, C. A. F.;
CAVALCANTI, R. N., BONA, E.; SILVA, M. A. A. P. Consumer acceptability and
purchase intent of probiotic yoghurt with added glucose oxidase using sensometrics,
98
artificial neural networks and logistic regression. International Journal of Dairy
Technology, v. 64, n 4, p. 549-556, 2011.
CRUZ, A. G., E. H. M. WALTER, R. S. CADENA, J. A. F. FARIA, H. M. A.
BOLINI AND A. M. FRATTINI FILETI. Monitoring the authenticity of low-fat yogurts by
an artificial neural network. Journal of Dairy Science.v. 92, n. 10, p. 4797-4804, 2009
DELWICHE, J. The impact of perceptual interactions on perceived flavor.
Food Quality and Preference, v. 15, p. 137-146, 2004.
DEMIATE, I. M.; KONKEL, F. E.; PEDROSO, R. A. Avaliação da qualidade
de amostras comerciais de doce de leite pastoso - composição química. Ciência e
Tecnologia de Alimentos, Campinas, v. 21, n. 1, p. 108-114, jan. 2001.
ERICH, S., ANZMANN, T., & FISCHER, L.. Quantification of lactose using
ion-pair RP-HPLC during enzymatic lactose hydrolysis of skim milk. Food Chemistry,
v. 135, p. 2393–2396, 2012
FELICIO, T. L., ESMERINO, E. A., CRUZ, A. G., NOGUEIRA, L. C.,
RAICES, R. S. L., DELIZA, R., BOLINI, H. M. A, &. POLLONIO, M. A. R. Cheese.
What is its contribution to the sodium intake of Brazilians? Appetite, v. 66, p. 84-88,
2013
FENNEMA, O. R. Food Chemistry. 3 ed. Nova Iorque: Marcel Dekker, 1996,
1069 p.
FERREIRA, A. M., BISLEVC, S. L., BENDIXENC, E., & ALMEIDA, A. M. .
The mammary gland in domestic ruminants: A systems biology perspective. Journal
of proteomics, v. 94, p. 110-123, 2013
GARCIA, K., S. SRIWATTANA, H. N. KYOON, J. A. H. CORREDOR, W.
PRINYAWIWATKUL.. Sensory Optimization of a Mayonnaise-Type Spread Made
with Rice Brain Oil and Soy Protein. Journal of Food Science. v. 74, p. 248-254, 2009
99
GARITTA, L., HOUGH, G., & SANCHEZ, R.. Sensory Shelf Life of Dulce de
Leche. Journal of Dairy Science, v. 87, p. 1601-1607, 2004
GIMÉNEZ, A., ARÉS, G., GAMBÁRO, C.. Consumer reaction to changes in
sensory profile of dulce de leche due to lactose hydrolysis. International Dairy
Journal, v. 18, p. 951–955, 2008
GOLON, A., & KUHNERT, N.. Unraveling the Chemical Composition of
Caramel. Journal of Agricultural and Food Chemistry, v. 60, p.3266−3274, 2012
GONZALEZ DE LLANO, D.; RODRIGUEZ, A.; CUESTA, P. Effect of lactic
starter cultures on the organic acid composition of milk and cheese during ripening-
analysis by HPLC. Journal of Applied Bacteriology, v. 80, p. 570-576, 1996.
GRANATO, D., CALADO, V. M. A., & JARVIS, B.. Observations on the use
of statistical methods in Food Science and Technology. Food Research International,
v. 55, p. 137–149, 2014
GUIMARÃES, Í. C. O., LEÃO, M. H. M. R., PIMENTA, C. J., FERREIRA, L.
O. & FERREIRA, E. B.. Development and description of light functional Dulce de
Leche with coffee. Ciênc. agrotec., v. 36, p. 195 -203, 2012
HOUGH, G., BRATCHELL, N., & WAKELING, I.. Consumer preference of
Dulce de Leche among students in the United Kingdom. Journal of Sensory Studies,
v. 7, p. 119-132, 1992a
HOUGH, G., BRATCHELL, N., & MACDOUGALL, D.. Sensory profiling of
Dulce de Leche, a dairy based confectionary product. Journal of Sensory Studies, 7,
157-178, 1992b.
IAL (Adolfo Lutz Institute) (2008). Métodos Físico-Químicos para Análise de
Alimentos (4th ed.). São Paulo, Brazil: Instituto Adolfo Lutz.
100
JIANG, Z., WANG, L., WU, W., WANG, Y. . Biological activities and
physicochemical properties of Maillard reaction products in sugar–bovine casein
peptide model systems. Food Chemistry, v. 141, p. 3837–3845, 2013
KONKEL, F. E., OLIVEIRA, S. M. R., SIMÕES, D. R. S., DEMIATE, I..M..
Avaliação sensorial de doce de leite pastoso com diferentes concentrações de
amido, Ciênc. Tecnol. Aliment.. v. 24, p. 249-254, 2004
KRISHNAMURTHY, R.; SRIVASTAVA, A. K.; PATON, J. E.; BELL, G. A.;
LEVY, D. C. Prediction of Consumer Liking from Trained Sensory Panel Information:
Evaluation of Neural Networks. Food Quality and Preference, v. 18, p. 275-285,
2007.
LEE, Y. S.; YOUM, G.; MEULLENET, J.F. Optimization of Consumer
Acceptability and Sensory Characteristics for Marinated Broiler Meat. Journal of Food
Science, v. 75, p. 478-484, 2011.
LLANO, D. G, RODRIGUEZ, A. & CUESTA, P.. Effect of lactic starter
cultures on the organic acid composition of milk and cheese during ripening-analysis
by HPLC, Journal of Applied Bacteriology. v. 80, p. 570-576, 1996
LUNA, M. P., & AGUILERA, J. M.. Kinetics of Colour Development of Molten
Glucose, Fructose and Sucrose at High Temperatures. Food Biophysics, v. 9, p. 61–
68, 2014
MACFIE, H. (2007). Consumer-led Food Product Development. Cambridge,
UK: Woodhead Publishing Limited,
MACFIE, H. J.; BRATCHELL, N.; GREENHOFF, K.; VALLIS, L. Designs to
Balance the Effect of Order of Presentation and First-order Carry-over Effects in Hall
Tests. Journal of Sensory Studies. V. 4, p. 129-148, 1989.
101
MDIC (Ministério do Desenvolvimento, Indústria e Comércio Exterior).
Sistema de Análise das Informações de Comércio Exterior, AliceWeb (2014). Url
aliceweb.desenvolvimento.gov.br. Accessed 05.04.14
MEULLENET, J. F.; XIONG, R.; FINDLAY, C. J. Multivariate and Probabilistic
Analyses of Sensory Science Problems. 5 ed. Iowa, US: Wiley-Blackwell, 2007. 256
p.
MOORE, J. C., SPINK, J., & LIPP, M.. Development and Application of a
Database of Food Ingredient Fraud and Economically Motivated Adulteration from
1980 to 2010. Journal of Food Science, v. 77, p. 118-126, 2012
MORO, O., & HOUGH, G.. Total Solids and Density Measurements of Dulce
de Leche, a Typical Argentine Dairy Product. Journal of Dairy Science, v. 68, p. 521-
525, 1985
MOSKOWITZ, H. R. 1983. Product Testing and Sensory Evaluation of Foods. Marketing and R & D Approaches. Food and Nutrition Press Inc., Connecticut, US.
NAES, T.; BROCKHOFF, P.; TOMIC, O. Statistics for Sensory and
Consumer Science. Chichester, West Sussex, UK: John Wiley & Sons, 2010, 282 p.
NAOZUKA, J., VIEIRA, E. C., NASCIMENTO, A. N., OLIVEIRA, P. V..
Elemental analysis of nuts and seeds by axially viewed ICP OES. Food Chemistry, v.
124, p. 1667-1672, 2011
OLIVEIRA, M. N., PENNA, A. L. B., & NEVAREZ, H. G. (2009); Production of
Evaporated Milk, Sweetened Condensed Milk and ‘Dulce de Leche. In A. Y. Tamime
(Eds.) Dairy Powders and Concentrated Products. (pp. 149-180). Oxford: Blackwell
Publishing.
PADILLA, B., BELLOCH, C., LÓPEZ-DÍEZ, J. J., FLORES, M., &
MANZANARES, P.. Potential impact of dairy yeasts on the typical flavour of
102
traditional ewes’ and goats’ cheeses. International Dairy Journal, v. 35, p. 122-129,
2014
PANSERI, S., SONCIN, S., CHIESA, L. M, BIONDI, P. A. . A headspace
solid-phase microextraction gas-chromatographic mass-spectrometric method (HS-
SPME–GC/MS) to quantify hexanal in butter during storage as marker of lipid
oxidation. Food Chemistry, v. 127, p. 886–889, 2011
PARAVISINI, L., GOURRAT-PERNIN, K., GOUTTEFANGEAS, C.,
MORETTON, C., NIGAY, H., DACREMONT, C., & GUICHARD, E. (2012).
Identification of compounds responsible for the odorant properties of aromatic
caramel. Flavour and Fragrance. Journal, v. 27,p. 424–432, 2012
PARC, A. L., LEE, H., CHEN, K., & BARILE, D... Rapid Quantification of
Functional Carbohydrates in Food Products. Food and Nutrition Sciences, v. 5, p.
71-78, 2014
PATHARE, P. B., OPARA, U. L., & AL-SAID, F. A... Colour Measurement
and Analysis in Fresh and Processed Foods: A Review. Food Bioprocess
Technology, v. 6, p. 36–60, 2013
PERRONE, I. T., STEPHANI, R.; NEVES, B. S. Doce de Leite: Aspectos
Tecnológicos. Juiz de Fora: Instituto de Laticínios Cândido Tostes, 2011. 186 p.
POTTER, N. N.; HOTCHKISS,J. H. Food Science. 5 ed. Nova Iorque:
Springuer Pub. Co, 1995, 531 p..
PRENTICE, R. L., PETTINGER, M. B., JACKSON, R. D., WACTAWSKI-
WENDE, J., LACROIX, A. Z., ANDERSON, G. L., CHLEBOWSKI, R. T., MANSON,
J. E., VAN HORN, L. VITOLINS, M. Z., DATTA, M., LEBLANC, E. S., CAULEY, J. A.,
& ROSSOUW, J. E. . Health risks and benefits from calcium and vitamin D
supplementation: Women's Health Initiative clinical trial and cohort study.
Osteoporosis International, v. 24, p. 567–580, 2013
103
RANALLI, N., ANDRÉS, S. C., & CALIFANO, A. N.. Physicochemical and
Rheological Characterization of “Dulce de Leche”. Journal of Texture Studies, v. 43,
p. 115-123, 2012
RODRÍGUEZ, A., PIAGENTINI, A., ROZYCKI, S., LEMA, P., PAULETTI, M.
S., & PANIZZOLO, L. A.. Color development in dulce de leche-like system.Effect of
heating time and pH. Revista de Laboratorio Tecnológico del Uruguay, p. 38-42,
2012
SAHIN, S. AND S. G. SUMNU. 2006. Physical Properties of Foods. Springer Science.
SILVA, M. J.; ALMEIDA, N. M.; SANTOS, J. G.; SANTOS, E. P.
Características Físico-químicas do Doce de Leite Pastoso Comercializado nos
Municípios de Bananeiras e Solânea-PB. In: JORNADA NACIONAL DA
AGROINDÚSTRIA, 2., 2007, Bananeiras.
SMIT, G. (2000). Dairy Processing – Improving Quality: Boca Raton, FL:
CRC Press.
STEINER, A. E., FOEGEDING, E. A., & DRAKE, M.. Descriptive analysis of
caramel texture. Journal of Sensory Studies, v. 18, p. 277-289, 2002.
STONE, H.; BLEIBAUM, R.N.; THOMAS, H.A. Sensory Evaluation Practices.
4 ed. San Diego: Academic Press, 2012. 438 p.
TANASE, C. M., GRIFFIN, P., KOSKI, K. G., COOPER, M. J., COCKELL, K.
A.. Sodium and potassium in composite food samples from the Canadian Total Diet
Study, Journal of Food Composition and Analysis.v. 24, p. 237–243, 2011.
VAN DEN DOOL, H., & KRATZ, P. D.. A generalization of the retention index
system including linear temperature programmed gas–liquid partition
chromatography. Journal of Chromatography, 11, 463–471, 1963
104
VARELA, P. AND G. ARES. 2012. Sensory Profiling, the Blurred Line
Between Sensory and Consumer Science. A Review of Novel Methods for Product
Characterization. Food Research International. v. 48, p. 893-908, 2012
VELASCO, V., QUEZADA, J., PARRA, C., CAMPOS, J., VILLALOBOS, R., &
WELLS, G). Effect of different l-carrageenan and sucrose concentrations on some
physical and sensory characteristics of solid-caramel-milk. Revista Chilena de
Ciencias Agropecuarias, v. 26, p. 25-35, 2010
VÉNICA, C. I., BERGAMINI, C. V., ZALAZAR, C. A., & PEROTTI, M. C..
Effect of Lactose Hydrolysis during Manufacture and Storage of Drinkable Yogurt.
Journal of Food & Nutritional Disorders,, 2013
VILANOVA, M.; CAMPO, E.; ESCUDERO, A.; GRAÑA, M.; MASAAND, A.;
CACHO, J. Volatile Composition and Sensory Properties of Vitisvinifera Red
Cultivars from North West Spain: Correlation Between Sensory and Instrumental
Analysis. Analytica Chimica Acta, v. 720, p. 104-111, 2012.
VISIOLI, F., & STRATA, A. Milk, Dairy Products, and Their Functional
Effects in Humans: A Narrative Review of Recent Evidence. American Society for
Nutrition, v. 5, p. 131–143, 2014
WADHWANI, R.; McMAHON, D.J. Color of low-fat cheese influences flavor
perception and consumer liking. Journal of Dairy Science, v. 95, n.5, p. 2336-2346,
2012
WHO (World Heath Organization).. Reducing salt intake in populations.
Paris: World Health Organization. 2006
ZABBIA, A., BUYS, E. M., & DE KOCK, H. L.. Undesirable Sulphur and
Carbonyl Flavor Compounds in UHT Milk: A Review. Critical Reviews in Food
Science and Nutrition, v. 52, p. 21-30, 2012
105
ZALAZAR, C. A.; PEROTTI, M. C.. Concentrated Dairy Products, Dulce de
Leche. In J. W. Fuquay, P. Fox, & P. McSweeney (Eds.), Enciclopedia of Dairy
Science (2 ed.) (pp. 874-880). Cornwall, UK: Academic Press. 2011
106
6 APÊNDICES
6.1 CONFIRMAÇÃO DE SUBMISSÃO DO ARTIGO INTITULADO: OPTIMIZATION OF THE DULCE DE LECHE PROCESSING: CONTRIBUTIONS OF THE SENSOMETRIC TECHNIQUES
107
6.2 CONFIRMAÇÃO DE SUBMISSÃO DO ARTIGO INTITULADO: DULCE DE LECHE, A TYPICAL PRODUCT OF LATIN AMERICA: CHARACTERIZATION BY PHYSICOCHEMICAL, OPTICAL AND INSTRUMENTAL METHODS